word_embeddings.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "colab": {
      "name": "word_embeddings.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lfoTYQOidI23"
      },
      "source": [
        "# Word embeddings\n",
        "\n",
        "The previous analyses have shown that fake news tends to use consistently inflammatory and subjective vocabulary, and tends to cover issues that may incite controversy.\n",
        "\n",
        "Let's drill down to the word level and look for connotations among words used in both fake and real news. This could reveal underlying biases that shape how certain words like `election` or `president` are perceived.\n",
        "\n",
        "1. [Load data](#Load-data)\n",
        "2. [Train embeddings](#Train-embedding-models)\n",
        "3. [Nearest neighbors](#Look-at-nearest-neighbors)\n",
        "4. [Different neighbors](#Find-words-with-different-neighbors)\n",
        "5. [Exploration](#Exploration)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "j9zcqQwKdI3K"
      },
      "source": [
        "### Load data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "djjre044dP5J",
        "outputId": "6c0362d2-7c5e-45c1-9d35-23835b4229d9"
      },
      "source": [
        "# NOTE: only run if you don't have data/ in Files\n",
        "!wget https://bitbucket.org/istewart6/core_tutorial_2020/raw/36e69f9d777319ae2cc94354cf57bd01f3e080b3/data.zip . ; unzip data.zip"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-12-09 19:49:23--  https://bitbucket.org/istewart6/core_tutorial_2020/raw/9feab1699d48005d9631f277771d91b2075d1256/data.zip\n",
            "Resolving bitbucket.org (bitbucket.org)... 104.192.141.1, 2406:da00:ff00::22e9:9f55, 2406:da00:ff00::22cd:e0db, ...\n",
            "Connecting to bitbucket.org (bitbucket.org)|104.192.141.1|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 230901661 (220M) [application/zip]\n",
            "Saving to: ‘data.zip’\n",
            "\n",
            "data.zip            100%[===================>] 220.20M  41.2MB/s    in 5.5s    \n",
            "\n",
            "2020-12-09 19:49:36 (39.9 MB/s) - ‘data.zip’ saved [230901661/230901661]\n",
            "\n",
            "--2020-12-09 19:49:36--  http://./\n",
            "Resolving . (.)... failed: No address associated with hostname.\n",
            "wget: unable to resolve host address ‘.’\n",
            "FINISHED --2020-12-09 19:49:36--\n",
            "Total wall clock time: 13s\n",
            "Downloaded: 1 files, 220M in 5.5s (39.9 MB/s)\n",
            "Archive:  data.zip\n",
            "   creating: data/\n",
            "   creating: data/fakeNewsDatasets/\n",
            "  inflating: data/fakeNewsDatasets/fake_news_small.tsv  \n",
            "  inflating: data/fakeNewsDatasets/real_news_small.tsv  \n",
            "   creating: data/fake_news_challenge/\n",
            "  inflating: data/fake_news_challenge/real_news_word2vec_embed.model  \n",
            "  inflating: data/fake_news_challenge/fake_news_glove_embed.model  \n",
            "  inflating: data/fake_news_challenge/Fake.csv  \n",
            "  inflating: data/fake_news_challenge/fake_news_word2vec_embed.model  \n",
            "  inflating: data/fake_news_challenge/real_news_glove_embed.model  \n",
            "  inflating: data/fake_news_challenge/True.csv  \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "I5n1PyGwndrr",
        "outputId": "bbabb31b-3e74-45a5-c80c-e2d8b3b36017"
      },
      "source": [
        "!pip install stop_words\n",
        "!pip install glove-python"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: stop_words in /usr/local/lib/python3.6/dist-packages (2018.7.23)\n",
            "Requirement already satisfied: glove-python in /usr/local/lib/python3.6/dist-packages (0.1.0)\n",
            "Requirement already satisfied: scipy in /usr/local/lib/python3.6/dist-packages (from glove-python) (1.4.1)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from glove-python) (1.18.5)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 206
        },
        "id": "txQCr1crdI3O",
        "outputId": "4f6b0941-ac53-4813-9fd7-1571e87346ba"
      },
      "source": [
        "## data = fake news challenge\n",
        "import pandas as pd\n",
        "fake_news_article_data = pd.read_csv('data/fake_news_challenge/Fake.csv', sep=',', index_col=False)\n",
        "real_news_article_data = pd.read_csv('data/fake_news_challenge/True.csv', sep=',', index_col=False)\n",
        "display(fake_news_article_data.head())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>title</th>\n",
              "      <th>text</th>\n",
              "      <th>subject</th>\n",
              "      <th>date</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Donald Trump Sends Out Embarrassing New Year’...</td>\n",
              "      <td>Donald Trump just couldn t wish all Americans ...</td>\n",
              "      <td>News</td>\n",
              "      <td>December 31, 2017</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Drunk Bragging Trump Staffer Started Russian ...</td>\n",
              "      <td>House Intelligence Committee Chairman Devin Nu...</td>\n",
              "      <td>News</td>\n",
              "      <td>December 31, 2017</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Sheriff David Clarke Becomes An Internet Joke...</td>\n",
              "      <td>On Friday, it was revealed that former Milwauk...</td>\n",
              "      <td>News</td>\n",
              "      <td>December 30, 2017</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Trump Is So Obsessed He Even Has Obama’s Name...</td>\n",
              "      <td>On Christmas day, Donald Trump announced that ...</td>\n",
              "      <td>News</td>\n",
              "      <td>December 29, 2017</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Pope Francis Just Called Out Donald Trump Dur...</td>\n",
              "      <td>Pope Francis used his annual Christmas Day mes...</td>\n",
              "      <td>News</td>\n",
              "      <td>December 25, 2017</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                               title  ...               date\n",
              "0   Donald Trump Sends Out Embarrassing New Year’...  ...  December 31, 2017\n",
              "1   Drunk Bragging Trump Staffer Started Russian ...  ...  December 31, 2017\n",
              "2   Sheriff David Clarke Becomes An Internet Joke...  ...  December 30, 2017\n",
              "3   Trump Is So Obsessed He Even Has Obama’s Name...  ...  December 29, 2017\n",
              "4   Pope Francis Just Called Out Donald Trump Dur...  ...  December 25, 2017\n",
              "\n",
              "[5 rows x 4 columns]"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M5xwjDpBdI3U"
      },
      "source": [
        "# ## clean data\n",
        "from nltk.tokenize import PunktSentenceTokenizer, WordPunctTokenizer\n",
        "sent_tokenizer = PunktSentenceTokenizer()\n",
        "word_tokenizer = WordPunctTokenizer()\n",
        "def get_sentence_word_tokens(text, word_tokenizer, sent_tokenizer):\n",
        "    text_sents = sent_tokenizer.tokenize(text)\n",
        "    text_sent_tokens = list(map(word_tokenizer.tokenize, text_sents))\n",
        "    return text_sent_tokens\n",
        "fake_news_sentences = fake_news_article_data.loc[:, 'text'].apply(lambda x: get_sentence_word_tokens(x, word_tokenizer, sent_tokenizer))\n",
        "real_news_sentences = real_news_article_data.loc[:, 'text'].apply(lambda x: get_sentence_word_tokens(x, word_tokenizer, sent_tokenizer))\n",
        "# flatten for processing\n",
        "from functools import reduce\n",
        "def flatten_list_data(data):\n",
        "    flat_data = []\n",
        "    for x in data:\n",
        "        flat_data.extend(x)\n",
        "    return flat_data\n",
        "fake_news_sentences = flatten_list_data(fake_news_sentences)\n",
        "real_news_sentences = flatten_list_data(real_news_sentences)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "B7K3Xau0dI3Y"
      },
      "source": [
        "### Train embedding models\n",
        "Let's first train the word embedding models on the full data."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wpIA-rCTdI3Z"
      },
      "source": [
        "## train word2vec embeddings\n",
        "from gensim.models.word2vec import Word2Vec\n",
        "def train_word2vec_model(text_sents, model_out_file):\n",
        "    dim = 50\n",
        "    alpha = 0.025\n",
        "    window = 5\n",
        "    min_count = 5\n",
        "    model = Word2Vec(sentences=text_sents, size=dim, alpha=alpha, window=window, min_count=min_count, seed=123)\n",
        "#     model.build_vocab(text_sents)\n",
        "    model.save(model_out_file)\n",
        "fake_news_word2vec_model_out_file = 'data/fake_news_challenge/fake_news_word2vec_embed.model'\n",
        "real_news_word2vec_model_out_file = 'data/fake_news_challenge/real_news_word2vec_embed.model'\n",
        "## training takes ~ 10 mins on Colab\n",
        "# SKIP these steps during tutorial to save time; use pre-trained model\n",
        "# train_word2vec_model(fake_news_sentences, fake_news_word2vec_model_out_file)\n",
        "# train_word2vec_model(real_news_sentences, real_news_word2vec_model_out_file)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2BD7hXQhdI3a"
      },
      "source": [
        "## load from file\n",
        "fake_news_word2vec_embed_model = Word2Vec.load(fake_news_word2vec_model_out_file)\n",
        "real_news_word2vec_embed_model = Word2Vec.load(real_news_word2vec_model_out_file)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5efiz4J0dI3b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "76716875-bd13-4ffc-e691-ab4dfb629599"
      },
      "source": [
        "## train Glove embeddings\n",
        "from glove import Glove, Corpus\n",
        "def fit_glove_model(text_sents, model_out_file):\n",
        "    dim = 50\n",
        "    learning_rate = 0.05\n",
        "    alpha = 0.025\n",
        "    random_state = 123\n",
        "    train_epochs = 100\n",
        "    num_threads = 4\n",
        "    window = 5\n",
        "    glove_corpus = Corpus()\n",
        "    glove_corpus.fit(text_sents, window=window)\n",
        "    glove_embed_model = Glove(no_components=dim, learning_rate=learning_rate, \n",
        "                              alpha=alpha, random_state=random_state)\n",
        "    # note: this takes ~ 5 minutes with 4 threads on a server\n",
        "    glove_embed_model.fit(glove_corpus.matrix, epochs=train_epochs,\n",
        "                          no_threads=num_threads, verbose=True)\n",
        "    glove_embed_model.add_dictionary(glove_corpus.dictionary)\n",
        "    glove_embed_model.save(model_out_file)\n",
        "fake_news_glove_model_out_file = 'data/fake_news_challenge/fake_news_glove_embed.model'\n",
        "real_news_glove_model_out_file = 'data/fake_news_challenge/real_news_glove_embed.model'\n",
        "## skipping these steps to save time during tutorial\n",
        "# fit_glove_model(fake_news_sentences, fake_news_glove_model_out_file)\n",
        "# fit_glove_model(real_news_sentences, real_news_glove_model_out_file)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Performing 100 training epochs with 4 threads\n",
            "Epoch 0\n",
            "Epoch 1\n",
            "Epoch 2\n",
            "Epoch 3\n",
            "Epoch 4\n",
            "Epoch 5\n",
            "Epoch 6\n",
            "Epoch 7\n",
            "Epoch 8\n",
            "Epoch 9\n",
            "Epoch 10\n",
            "Epoch 11\n",
            "Epoch 12\n",
            "Epoch 13\n",
            "Epoch 14\n",
            "Epoch 15\n",
            "Epoch 16\n",
            "Epoch 17\n",
            "Epoch 18\n",
            "Epoch 19\n",
            "Epoch 20\n",
            "Epoch 21\n",
            "Epoch 22\n",
            "Epoch 23\n",
            "Epoch 24\n",
            "Epoch 25\n",
            "Epoch 26\n",
            "Epoch 27\n",
            "Epoch 28\n",
            "Epoch 29\n",
            "Epoch 30\n",
            "Epoch 31\n",
            "Epoch 32\n",
            "Epoch 33\n",
            "Epoch 34\n",
            "Epoch 35\n",
            "Epoch 36\n",
            "Epoch 37\n",
            "Epoch 38\n",
            "Epoch 39\n",
            "Epoch 40\n",
            "Epoch 41\n",
            "Epoch 42\n",
            "Epoch 43\n",
            "Epoch 44\n",
            "Epoch 45\n",
            "Epoch 46\n",
            "Epoch 47\n",
            "Epoch 48\n",
            "Epoch 49\n",
            "Epoch 50\n",
            "Epoch 51\n",
            "Epoch 52\n",
            "Epoch 53\n",
            "Epoch 54\n",
            "Epoch 55\n",
            "Epoch 56\n",
            "Epoch 57\n",
            "Epoch 58\n",
            "Epoch 59\n",
            "Epoch 60\n",
            "Epoch 61\n",
            "Epoch 62\n",
            "Epoch 63\n",
            "Epoch 64\n",
            "Epoch 65\n",
            "Epoch 66\n",
            "Epoch 67\n",
            "Epoch 68\n",
            "Epoch 69\n",
            "Epoch 70\n",
            "Epoch 71\n",
            "Epoch 72\n",
            "Epoch 73\n",
            "Epoch 74\n",
            "Epoch 75\n",
            "Epoch 76\n",
            "Epoch 77\n",
            "Epoch 78\n",
            "Epoch 79\n",
            "Epoch 80\n",
            "Epoch 81\n",
            "Epoch 82\n",
            "Epoch 83\n",
            "Epoch 84\n",
            "Epoch 85\n",
            "Epoch 86\n",
            "Epoch 87\n",
            "Epoch 88\n",
            "Epoch 89\n",
            "Epoch 90\n",
            "Epoch 91\n",
            "Epoch 92\n",
            "Epoch 93\n",
            "Epoch 94\n",
            "Epoch 95\n",
            "Epoch 96\n",
            "Epoch 97\n",
            "Epoch 98\n",
            "Epoch 99\n",
            "Performing 100 training epochs with 4 threads\n",
            "Epoch 0\n",
            "Epoch 1\n",
            "Epoch 2\n",
            "Epoch 3\n",
            "Epoch 4\n",
            "Epoch 5\n",
            "Epoch 6\n",
            "Epoch 7\n",
            "Epoch 8\n",
            "Epoch 9\n",
            "Epoch 10\n",
            "Epoch 11\n",
            "Epoch 12\n",
            "Epoch 13\n",
            "Epoch 14\n",
            "Epoch 15\n",
            "Epoch 16\n",
            "Epoch 17\n",
            "Epoch 18\n",
            "Epoch 19\n",
            "Epoch 20\n",
            "Epoch 21\n",
            "Epoch 22\n",
            "Epoch 23\n",
            "Epoch 24\n",
            "Epoch 25\n",
            "Epoch 26\n",
            "Epoch 27\n",
            "Epoch 28\n",
            "Epoch 29\n",
            "Epoch 30\n",
            "Epoch 31\n",
            "Epoch 32\n",
            "Epoch 33\n",
            "Epoch 34\n",
            "Epoch 35\n",
            "Epoch 36\n",
            "Epoch 37\n",
            "Epoch 38\n",
            "Epoch 39\n",
            "Epoch 40\n",
            "Epoch 41\n",
            "Epoch 42\n",
            "Epoch 43\n",
            "Epoch 44\n",
            "Epoch 45\n",
            "Epoch 46\n",
            "Epoch 47\n",
            "Epoch 48\n",
            "Epoch 49\n",
            "Epoch 50\n",
            "Epoch 51\n",
            "Epoch 52\n",
            "Epoch 53\n",
            "Epoch 54\n",
            "Epoch 55\n",
            "Epoch 56\n",
            "Epoch 57\n",
            "Epoch 58\n",
            "Epoch 59\n",
            "Epoch 60\n",
            "Epoch 61\n",
            "Epoch 62\n",
            "Epoch 63\n",
            "Epoch 64\n",
            "Epoch 65\n",
            "Epoch 66\n",
            "Epoch 67\n",
            "Epoch 68\n",
            "Epoch 69\n",
            "Epoch 70\n",
            "Epoch 71\n",
            "Epoch 72\n",
            "Epoch 73\n",
            "Epoch 74\n",
            "Epoch 75\n",
            "Epoch 76\n",
            "Epoch 77\n",
            "Epoch 78\n",
            "Epoch 79\n",
            "Epoch 80\n",
            "Epoch 81\n",
            "Epoch 82\n",
            "Epoch 83\n",
            "Epoch 84\n",
            "Epoch 85\n",
            "Epoch 86\n",
            "Epoch 87\n",
            "Epoch 88\n",
            "Epoch 89\n",
            "Epoch 90\n",
            "Epoch 91\n",
            "Epoch 92\n",
            "Epoch 93\n",
            "Epoch 94\n",
            "Epoch 95\n",
            "Epoch 96\n",
            "Epoch 97\n",
            "Epoch 98\n",
            "Epoch 99\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PVWvcfObdI3c"
      },
      "source": [
        "## reload models after training\n",
        "fake_news_glove_embed_model = Glove.load(fake_news_glove_model_out_file)\n",
        "real_news_glove_embed_model = Glove.load(real_news_glove_model_out_file)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "T2HWweTNOt6U",
        "outputId": "8b441e73-6d6b-4a3b-8bce-efaa3fabe9b7"
      },
      "source": [
        "# !rm data.zip\n",
        "!zip -r data.zip data/"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  adding: data/ (stored 0%)\n",
            "  adding: data/fakeNewsDatasets/ (stored 0%)\n",
            "  adding: data/fakeNewsDatasets/fake_news_small.tsv (deflated 63%)\n",
            "  adding: data/fakeNewsDatasets/real_news_small.tsv (deflated 58%)\n",
            "  adding: data/fake_news_challenge/ (stored 0%)\n",
            "  adding: data/fake_news_challenge/fake_news_glove_embed.model (deflated 6%)\n",
            "  adding: data/fake_news_challenge/real_news_word2vec_embed.model (deflated 30%)\n",
            "  adding: data/fake_news_challenge/Fake.csv (deflated 62%)\n",
            "  adding: data/fake_news_challenge/fake_news_word2vec_embed.model (deflated 29%)\n",
            "  adding: data/fake_news_challenge/True.csv (deflated 65%)\n",
            "  adding: data/fake_news_challenge/real_news_glove_embed.model (deflated 6%)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9IOi7sX-dI3d"
      },
      "source": [
        "### Look at nearest neighbors"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6GPn73PQdI3d"
      },
      "source": [
        "Let's start out by looking at the nearest neighbors for some test words. \n",
        "\n",
        "We'll get the test words by filtering from the most frequent words."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BvyW7r7VdI3i",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "f4c9d5ab-04d5-40f2-8234-a00f43ac89fb"
      },
      "source": [
        "from collections import Counter\n",
        "from stop_words import get_stop_words\n",
        "import pandas as pd\n",
        "pd.set_option('display.max_rows', 100)\n",
        "news_word_counter = Counter()\n",
        "for sent_i in fake_news_sentences:\n",
        "    news_word_counter.update(sent_i)\n",
        "for sent_i in real_news_sentences:\n",
        "    news_word_counter.update(sent_i)\n",
        "news_word_counts = pd.Series(dict(news_word_counter)).sort_values(inplace=False, ascending=False)\n",
        "en_stops = set(get_stop_words('en')) & set(news_word_counts.index)\n",
        "news_word_counts.drop(en_stops, inplace=True)\n",
        "display(news_word_counts.head(100))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              ".                 899948\n",
              ",                 872906\n",
              "s                 231123\n",
              "-                 191679\n",
              "Trump             132996\n",
              "said              132673\n",
              "The               115553\n",
              "’                  70098\n",
              ":                  63415\n",
              ")                  63128\n",
              "I                  62061\n",
              "(                  57607\n",
              "U                  54124\n",
              "“                  53931\n",
              "t                  53069\n",
              "S                  51453\n",
              "will               48561\n",
              "people             39640\n",
              "President          36123\n",
              "one                32627\n",
              "also               30519\n",
              "It                 29705\n",
              "Reuters            29343\n",
              "Clinton            28556\n",
              "Donald             28066\n",
              "Obama              28035\n",
              "?                  27492\n",
              "government         26786\n",
              "can                26401\n",
              "He                 26344\n",
              "Republican         25450\n",
              "House              25418\n",
              "In                 25292\n",
              "year               24425\n",
              ",”                 24296\n",
              "/                  23695\n",
              "told               23330\n",
              "United             22860\n",
              "just               22835\n",
              "We                 22082\n",
              "like               21830\n",
              "state              21276\n",
              "campaign           21257\n",
              "time               20355\n",
              "A                  19937\n",
              "election           19882\n",
              "two                19850\n",
              "president          19578\n",
              "States             19182\n",
              "last               18973\n",
              "But                18857\n",
              "$                  18357\n",
              "country            17936\n",
              "(@                 17884\n",
              "now                17279\n",
              "This               17223\n",
              "new                16820\n",
              "Hillary            16757\n",
              "years              16552\n",
              "former             16440\n",
              "re                 16433\n",
              "first              16337\n",
              "White              16161\n",
              "American           16132\n",
              "even               15892\n",
              "!                  15526\n",
              "made               15432\n",
              "media              15262\n",
              "say                15240\n",
              "political          14885\n",
              ".”                 14755\n",
              "many               14583\n",
              "get                14457\n",
              "law                14442\n",
              "percent            14392\n",
              "Republicans        14109\n",
              "going              14030\n",
              "Russia             13943\n",
              "make               13937\n",
              "right              13682\n",
              "America            13490\n",
              "back               13477\n",
              "party              13365\n",
              "New                13346\n",
              "called             13170\n",
              "State              13109\n",
              "presidential       13060\n",
              "week               13048\n",
              "since              12896\n",
              "administration     12818\n",
              "support            12788\n",
              "And                12766\n",
              "Senate             12706\n",
              "including          12586\n",
              "know               12491\n",
              "vote               12442\n",
              "way                12440\n",
              "Democratic         12354\n",
              "police             12347\n",
              "think              12269\n",
              "dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QgqR7Sc6dI3k"
      },
      "source": [
        "test_words = ['Trump', 'President', 'election', 'Republicans', 'Democratic']"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XldaE1ZCdI3k",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "111eaff0-0e22-4aaf-8555-1c3008511088"
      },
      "source": [
        "## test word2vec first\n",
        "N_neighbors = 10\n",
        "for test_word_i in test_words:\n",
        "    print(f'testing word = {test_word_i}')\n",
        "    print(f'\\tfake news neighbors')\n",
        "    print(fake_news_word2vec_embed_model.wv.most_similar(test_word_i, topn=N_neighbors))\n",
        "    print(f'\\treal news neighbors')\n",
        "    print(real_news_word2vec_embed_model.wv.most_similar(test_word_i, topn=N_neighbors))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "testing word = Trump\n",
            "\tfake news neighbors\n",
            "[('Cruz', 0.7542868852615356), ('Obama', 0.7326431274414062), ('he', 0.6627222299575806), ('Rubio', 0.6566429138183594), ('trump', 0.6353479623794556), ('GOP', 0.6161357164382935), ('him', 0.6154471039772034), ('He', 0.6065627336502075), ('Putin', 0.5974994897842407), ('Bush', 0.5890234112739563)]\n",
            "\treal news neighbors\n",
            "[('him', 0.6888852119445801), ('Obama', 0.6868149638175964), ('Macron', 0.6779283285140991), ('Cruz', 0.6577001214027405), ('Rubio', 0.6564674973487854), ('Clinton', 0.6557707786560059), ('Abe', 0.651238739490509), ('he', 0.6333320140838623), ('Pence', 0.6211468577384949), ('Tillerson', 0.6183730363845825)]\n",
            "testing word = President\n",
            "\tfake news neighbors\n",
            "[('president', 0.8004584312438965), ('Barack', 0.7969117760658264), ('administration', 0.7702555656433105), ('presidency', 0.7402648329734802), ('Administration', 0.7270553112030029), ('Michelle', 0.616836667060852), ('nominee', 0.6052952408790588), ('regime', 0.5850890874862671), ('Presidency', 0.576806366443634), ('Donald', 0.5613030195236206)]\n",
            "\treal news neighbors\n",
            "[('president', 0.6308104395866394), ('predecessor', 0.570038914680481), ('Tusk', 0.5458980798721313), ('Jr', 0.5286433696746826), ('presidency', 0.5145781636238098), ('Rumsfeld', 0.5104163885116577), ('counterpart', 0.491938054561615), ('Governor', 0.4909416437149048), ('Maia', 0.48573383688926697), ('leader', 0.48039546608924866)]\n",
            "testing word = election\n",
            "\tfake news neighbors\n",
            "[('elections', 0.778891921043396), ('primaries', 0.7418304085731506), ('race', 0.7160807251930237), ('nomination', 0.7149498462677002), ('primary', 0.7122625112533569), ('contest', 0.691138744354248), ('presidential', 0.6808596849441528), ('presidency', 0.6701884865760803), ('candidate', 0.6683717370033264), ('polls', 0.6597499251365662)]\n",
            "\treal news neighbors\n",
            "[('elections', 0.8428527116775513), ('race', 0.7417063117027283), ('contest', 0.6856149435043335), ('electoral', 0.6704519987106323), ('nominating', 0.646630585193634), ('vote', 0.6321263909339905), ('ballot', 0.631411612033844), ('campaign', 0.6264020800590515), ('candidate', 0.6156679391860962), ('party', 0.6144907474517822)]\n",
            "testing word = Republicans\n",
            "\tfake news neighbors\n",
            "[('Democrats', 0.9445343017578125), ('conservatives', 0.8779096007347107), ('Americans', 0.726625919342041), ('lawmakers', 0.7197590470314026), ('politicians', 0.7077614068984985), ('voters', 0.7074059844017029), ('progressives', 0.6943092942237854), ('candidates', 0.6897746324539185), ('liberals', 0.6881696581840515), ('GOP', 0.6816704273223877)]\n",
            "\treal news neighbors\n",
            "[('Democrats', 0.9397876262664795), ('conservatives', 0.8157765865325928), ('lawmakers', 0.7968851327896118), ('parties', 0.7209782004356384), ('Lawmakers', 0.7072699666023254), ('voters', 0.6853541135787964), ('senators', 0.6793911457061768), ('Conservatives', 0.6720776557922363), ('candidates', 0.665237545967102), ('supporters', 0.645807147026062)]\n",
            "testing word = Democratic\n",
            "\tfake news neighbors\n",
            "[('Republican', 0.8832844495773315), ('Democrat', 0.8314604759216309), ('GOP', 0.7758756279945374), ('Tea', 0.7429810762405396), ('republican', 0.6818928122520447), ('Libertarian', 0.6424157619476318), ('Communist', 0.6408422589302063), ('Labour', 0.622765839099884), ('presidential', 0.5980229377746582), ('Senate', 0.5854494571685791)]\n",
            "\treal news neighbors\n",
            "[('Republican', 0.8854277729988098), ('Conservative', 0.7831263542175293), ('Socialist', 0.7594935894012451), ('PRI', 0.7350648045539856), ('Labour', 0.7280474901199341), ('ANC', 0.7235405445098877), ('Tea', 0.7129971385002136), ('AK', 0.6949486136436462), ('Green', 0.6831214427947998), ('Regions', 0.6611542701721191)]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P8kby2VZdI3l"
      },
      "source": [
        "For `word2vec`:\n",
        "- `Trump` is associated with almost exclusively Republican politicians (and Putin) in fake news and with a mix of international politicians in real news\n",
        "- `President` is associated more with U.S. politics in fake news and more with international politicians in real news\n",
        "- `Democratic` are associated more with U.S. politics in fake news and more with general political parties in real news"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_VjkbTOMdI3m",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1b6d15f6-a313-49fc-8d54-08bb9560d635"
      },
      "source": [
        "## test Glove embeddings\n",
        "N_neighbors = 10\n",
        "for test_word_i in test_words:\n",
        "    print(f'testing word = {test_word_i}')\n",
        "    print(f'\\tfake news neighbors')\n",
        "    print(fake_news_glove_embed_model.most_similar(test_word_i, number=N_neighbors))\n",
        "    print(f'\\treal news neighbors')\n",
        "    print(real_news_glove_embed_model.most_similar(test_word_i, number=N_neighbors))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "testing word = Trump\n",
            "\tfake news neighbors\n",
            "[('Donald', 0.8995104048828402), ('he', 0.7510152023913843), ('President', 0.7289551160683528), ('elect', 0.7018952339996948), ('his', 0.6894683958666277), ('him', 0.6590607131669851), ('supporter', 0.650652397522237), ('Obama', 0.6459602087651001), ('presidency', 0.6456801048266373)]\n",
            "\treal news neighbors\n",
            "[('Donald', 0.8999952077484253), ('Obama', 0.740171056813851), ('Clinton', 0.7259076263176928), ('he', 0.7161130616849752), ('Putin', 0.7157221029232305), ('president', 0.7005582815167459), ('administration', 0.699648394966008), ('elect', 0.6946826981167531), ('He', 0.6930951709096086)]\n",
            "testing word = President\n",
            "\tfake news neighbors\n",
            "[('Obama', 0.8787505866253105), ('president', 0.8297460406112078), ('Barack', 0.8097643017568634), ('Donald', 0.7786599002096065), ('ObamaE', 0.7675834296393814), ('administration', 0.7636917647469326), ('elect', 0.7305534045344345), ('Trump', 0.728955116068353), ('Putin', 0.6716257742337292)]\n",
            "\treal news neighbors\n",
            "[('Barack', 0.8069133844365693), ('Obama', 0.7652734865579703), ('Xi', 0.7325036014727178), ('elect', 0.7311648468273894), ('president', 0.7286690629279299), ('Donald', 0.7212482888035131), ('Vice', 0.71201805518998), ('Vladimir', 0.6740482606774978), ('Emmanuel', 0.6668299198952441)]\n",
            "testing word = election\n",
            "\tfake news neighbors\n",
            "[('presidential', 0.7811715484475453), ('win', 0.7261690813918633), ('won', 0.7110077429433272), ('primary', 0.6971328660966708), ('campaign', 0.6942696323064391), ('elections', 0.6932626481303165), ('2016', 0.6846305856893837), ('race', 0.677450156308952), ('process', 0.6646657418153342)]\n",
            "\treal news neighbors\n",
            "[('2016', 0.8256332350096169), ('presidential', 0.8030988071987903), ('elections', 0.7974120117665089), ('November', 0.7798220618044112), ('July', 0.7010202539719728), ('race', 0.6840903959109801), ('vote', 0.6827562229789786), ('2018', 0.6754497573996968), ('June', 0.6733627378548077)]\n",
            "testing word = Republicans\n",
            "\tfake news neighbors\n",
            "[('Democrats', 0.8961469250952033), ('conservatives', 0.8249306627989923), ('GOP', 0.743651580451974), ('Congress', 0.7414049134419138), ('Republican', 0.7348188380111701), ('won', 0.7147056210021069), ('win', 0.6797482683829427), ('voters', 0.6738124422611418), ('still', 0.6723307985608485)]\n",
            "\treal news neighbors\n",
            "[('Democrats', 0.9045199734269047), ('lawmakers', 0.8426598839319285), ('both', 0.7720973944454675), ('Congress', 0.7409623018801241), ('senators', 0.7358072893656712), ('Republican', 0.7017435893125257), ('Senate', 0.6865402784244785), ('some', 0.6593886388500072), ('opposed', 0.6590586549145524)]\n",
            "testing word = Democratic\n",
            "\tfake news neighbors\n",
            "[('Republican', 0.8744266278592752), ('Democrat', 0.8326532449964446), ('Party', 0.8207860693817824), ('GOP', 0.8101995298698995), ('presidential', 0.8035585336107657), ('candidates', 0.7607017076514181), ('candidate', 0.7495126222127931), ('primary', 0.7350840211891567), ('party', 0.7074662987574949)]\n",
            "\treal news neighbors\n",
            "[('Republican', 0.8228312220800497), ('candidate', 0.7685453561919888), ('Democrat', 0.761564037514429), ('Hillary', 0.7299404655175027), ('nominee', 0.7055108716850624), ('Party', 0.6879932813496151), ('candidates', 0.6853048236783968), ('presumptive', 0.682881983471366), ('rival', 0.6819931193872162)]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mqFxUXcjdI3n"
      },
      "source": [
        "For `Glove`:\n",
        "- `Trump` is associated with himself (and news network? `Q13FOXWATCH`) in fake news and with other presidents in real news\n",
        "- `President` is associated with Trump and Obama in fake news and more with international politicians in real news\n",
        "- `Democratic` is associated with U.S. party politics in both fake and real news\n",
        "\n",
        "This qualitative analysis helps us understand that some words may indeed have significant divergence in their connotations between the different data sets, while others are more stable."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jk8l_dztdI3n"
      },
      "source": [
        "### Find words with different neighbors"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o7HujsI_dI3o"
      },
      "source": [
        "Which words are the most different across the data?\n",
        "\n",
        "We'll measure \"difference\" using the overlap in nearest neighbors (i.e. Jaccard similarity).\n",
        "\n",
        "$$\\text{diff(word1, word2)} = 1 - \\frac{\\text{neighbors(word1)} \\: \\cap \\: \\text{neighbors(word2)}}{\\text{neighbors(word1)} \\cup \\text{neighbors(word2)}}$$\n",
        "\n",
        "A difference of 100% means that the words have no neighbors in common, while a difference of 0% means that the words have identical neighbors."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "87cUxSkIdI3o"
      },
      "source": [
        "def compute_neighbor_diff(neighbors_1, neighbors_2):\n",
        "    neighbor_intersect = set(neighbors_1) & set(neighbors_2)\n",
        "    neighbor_union = set(neighbors_1) | set(neighbors_2)\n",
        "    neighbor_diff = 1 - len(neighbor_intersect) / len(neighbor_union)\n",
        "    return neighbor_diff\n",
        "def compute_neighbor_diff_model(word, model_1, model_2, N_neighbor, model_type='word2vec'):\n",
        "    if(model_type == 'word2vec'):\n",
        "        neighbors_1, neighbor_scores_1 = zip(*model_1.wv.most_similar(word, topn=N_neighbor))\n",
        "        neighbors_2, neighbor_scores_2 = zip(*model_2.wv.most_similar(word, topn=N_neighbor))\n",
        "    elif(model_type == 'glove'):\n",
        "        neighbors_1, neighbor_scores_1 = zip(*model_1.most_similar(word, number=N_neighbor))\n",
        "        neighbors_2, neighbor_scores_2 = zip(*model_2.most_similar(word, number=N_neighbor))\n",
        "    neighbor_diff = compute_neighbor_diff(neighbors_1, neighbors_2)\n",
        "    return neighbor_diff"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NSs14U8bdI3p",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "bcce82d6-a81f-4e7d-8d34-1bfac0ce2b86"
      },
      "source": [
        "# get shared vocabulary\n",
        "shared_word2vec_vocab = list(set(fake_news_word2vec_embed_model.wv.vocab.keys()) & set(real_news_word2vec_embed_model.wv.vocab.keys()))\n",
        "print(f'{len(shared_word2vec_vocab)} words in word2vec vocab')\n",
        "# compute neighbor differences for all valid words\n",
        "model_type = 'word2vec'\n",
        "N_neighbor = 10\n",
        "fake_vs_real_word2vec_neighbor_diffs = list(map(lambda x: compute_neighbor_diff_model(x, fake_news_word2vec_embed_model, real_news_word2vec_embed_model, N_neighbor, model_type=model_type), shared_word2vec_vocab))\n",
        "# add vocabulary as index\n",
        "fake_vs_real_word2vec_neighbor_diffs = pd.Series(fake_vs_real_word2vec_neighbor_diffs, index=shared_word2vec_vocab)\n",
        "fake_vs_real_word2vec_neighbor_diffs.sort_values(inplace=True, ascending=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "23922 words in word2vec vocab\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "VxYywzvUdI3q",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "08d3482a-79c0-4af7-bc4c-82296d74769f"
      },
      "source": [
        "top_k = 20\n",
        "print('words with most neighbor difference')\n",
        "print(fake_vs_real_word2vec_neighbor_diffs.head(top_k))\n",
        "print('words with most neighbor similarity')\n",
        "print(fake_vs_real_word2vec_neighbor_diffs.tail(top_k))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "words with most neighbor difference\n",
            "swath           1.0\n",
            "Missing         1.0\n",
            "WIN             1.0\n",
            "Policies        1.0\n",
            "Vasquez         1.0\n",
            "transitioned    1.0\n",
            "wheelchairs     1.0\n",
            "reprieve        1.0\n",
            "testing         1.0\n",
            "Township        1.0\n",
            "Patriot         1.0\n",
            "Players         1.0\n",
            "rough           1.0\n",
            "facade          1.0\n",
            "referral        1.0\n",
            "106             1.0\n",
            "Nike            1.0\n",
            "psyche          1.0\n",
            "WHO             1.0\n",
            "tentatively     1.0\n",
            "dtype: float64\n",
            "words with most neighbor similarity\n",
            "March        0.181818\n",
            "aside        0.181818\n",
            "April        0.181818\n",
            "his          0.181818\n",
            "February     0.181818\n",
            "June         0.181818\n",
            "July         0.181818\n",
            "anyone       0.181818\n",
            "September    0.181818\n",
            "December     0.181818\n",
            "Tuesday      0.181818\n",
            "cannot       0.181818\n",
            "November     0.181818\n",
            "two          0.181818\n",
            "January      0.181818\n",
            "He           0.181818\n",
            "Thursday     0.181818\n",
            "Friday       0.181818\n",
            "Wednesday    0.181818\n",
            "Monday       0.000000\n",
            "dtype: float64\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YfDnQgWfdI3q"
      },
      "source": [
        "The words with the biggest neighbor differences don't seem to be super informative and may reflect topical differences (e.g. fake news tends to discuss `Patriot` more often and therefore has more consistent nearest neighbors).\n",
        "\n",
        "What if we restrict to the most frequent words?"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "ONUjIEqhdI3r",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "8430bf0c-0a84-4101-e8c3-16a713743773"
      },
      "source": [
        "# filter the word counts to include words that are in the shared vocab\n",
        "word2vec_vocab_news_word_counts = news_word_counts.loc[(news_word_counts.index & set(shared_word2vec_vocab))]\n",
        "word2vec_neighbor_diff_count_data = pd.concat([fake_vs_real_word2vec_neighbor_diffs, \n",
        "                                               word2vec_vocab_news_word_counts], axis=1).dropna()\n",
        "word2vec_neighbor_diff_count_data.columns = ['neighbor_diff', 'count']\n",
        "# sort data by difference, then frequency\n",
        "word2vec_neighbor_diff_count_data.sort_values(['neighbor_diff', 'count'], inplace=True, ascending=False)\n",
        "top_k = 100\n",
        "display(word2vec_neighbor_diff_count_data.head(top_k))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>neighbor_diff</th>\n",
              "      <th>count</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>(</th>\n",
              "      <td>1.0</td>\n",
              "      <td>57607.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>U</th>\n",
              "      <td>1.0</td>\n",
              "      <td>54124.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Donald</th>\n",
              "      <td>1.0</td>\n",
              "      <td>28066.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>/</th>\n",
              "      <td>1.0</td>\n",
              "      <td>23695.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>United</th>\n",
              "      <td>1.0</td>\n",
              "      <td>22860.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>New</th>\n",
              "      <td>1.0</td>\n",
              "      <td>13346.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>news</th>\n",
              "      <td>1.0</td>\n",
              "      <td>10343.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>via</th>\n",
              "      <td>1.0</td>\n",
              "      <td>9650.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>News</th>\n",
              "      <td>1.0</td>\n",
              "      <td>9025.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>image</th>\n",
              "      <td>1.0</td>\n",
              "      <td>8719.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>com</th>\n",
              "      <td>1.0</td>\n",
              "      <td>8696.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>foreign</th>\n",
              "      <td>1.0</td>\n",
              "      <td>8507.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>show</th>\n",
              "      <td>1.0</td>\n",
              "      <td>7552.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>US</th>\n",
              "      <td>1.0</td>\n",
              "      <td>7472.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>;</th>\n",
              "      <td>1.0</td>\n",
              "      <td>7395.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>twitter</th>\n",
              "      <td>1.0</td>\n",
              "      <td>6796.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>&amp;</th>\n",
              "      <td>1.0</td>\n",
              "      <td>5817.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>[</th>\n",
              "      <td>1.0</td>\n",
              "      <td>5781.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2017</th>\n",
              "      <td>1.0</td>\n",
              "      <td>5635.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>'</th>\n",
              "      <td>1.0</td>\n",
              "      <td>5238.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>head</th>\n",
              "      <td>1.0</td>\n",
              "      <td>5078.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>yet</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4922.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>co</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4916.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>gun</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4894.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>talks</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4751.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>process</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4734.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>face</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4714.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>General</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4702.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Mr</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4537.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>matter</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4456.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>GOP</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4433.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>No</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4283.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Bill</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4141.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Images</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4073.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Watch</th>\n",
              "      <td>1.0</td>\n",
              "      <td>4071.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>*</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3985.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>due</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3875.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>defense</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3847.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>force</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3761.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>special</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3681.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>post</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3638.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>immediately</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3393.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3335.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>capital</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3325.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>air</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3126.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>J</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3095.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>=</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3053.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>enforcement</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3021.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>host</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3021.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>First</th>\n",
              "      <td>1.0</td>\n",
              "      <td>3009.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>hit</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2872.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>parliament</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2805.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>City</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2797.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>.-</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2793.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Bernie</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2745.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>hand</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2738.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>O</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2730.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>watch</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2718.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>record</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2670.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>independence</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2473.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>West</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2427.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>reality</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2424.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>fear</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2386.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>read</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2329.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Century</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2302.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>corruption</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2263.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Johnson</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2253.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>popular</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2234.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>room</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2230.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>view</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2213.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>status</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2184.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>R</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2179.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>form</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2157.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>People</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2121.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Brexit</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2117.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>opinion</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2100.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>base</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2099.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>share</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2080.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Wire</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2071.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>numbers</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2056.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>terms</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2044.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>coverage</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2016.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>operations</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2012.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Foundation</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2008.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Of</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2003.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Just</th>\n",
              "      <td>1.0</td>\n",
              "      <td>2001.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Joe</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1997.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Kelly</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1989.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>cover</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1973.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>poor</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1966.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>favor</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1944.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>body</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1916.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>standing</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1914.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>agree</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1903.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Paris</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1885.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>test</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1866.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>individual</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1824.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>screen</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1819.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Sen</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1804.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>referendum</th>\n",
              "      <td>1.0</td>\n",
              "      <td>1791.0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "              neighbor_diff    count\n",
              "(                       1.0  57607.0\n",
              "U                       1.0  54124.0\n",
              "Donald                  1.0  28066.0\n",
              "/                       1.0  23695.0\n",
              "United                  1.0  22860.0\n",
              "New                     1.0  13346.0\n",
              "news                    1.0  10343.0\n",
              "via                     1.0   9650.0\n",
              "News                    1.0   9025.0\n",
              "image                   1.0   8719.0\n",
              "com                     1.0   8696.0\n",
              "foreign                 1.0   8507.0\n",
              "show                    1.0   7552.0\n",
              "US                      1.0   7472.0\n",
              ";                       1.0   7395.0\n",
              "twitter                 1.0   6796.0\n",
              "&                       1.0   5817.0\n",
              "[                       1.0   5781.0\n",
              "2017                    1.0   5635.0\n",
              "'                       1.0   5238.0\n",
              "head                    1.0   5078.0\n",
              "yet                     1.0   4922.0\n",
              "co                      1.0   4916.0\n",
              "gun                     1.0   4894.0\n",
              "talks                   1.0   4751.0\n",
              "process                 1.0   4734.0\n",
              "face                    1.0   4714.0\n",
              "General                 1.0   4702.0\n",
              "Mr                      1.0   4537.0\n",
              "matter                  1.0   4456.0\n",
              "GOP                     1.0   4433.0\n",
              "No                      1.0   4283.0\n",
              "Bill                    1.0   4141.0\n",
              "Images                  1.0   4073.0\n",
              "Watch                   1.0   4071.0\n",
              "*                       1.0   3985.0\n",
              "due                     1.0   3875.0\n",
              "defense                 1.0   3847.0\n",
              "force                   1.0   3761.0\n",
              "special                 1.0   3681.0\n",
              "post                    1.0   3638.0\n",
              "immediately             1.0   3393.0\n",
              "...                     1.0   3335.0\n",
              "capital                 1.0   3325.0\n",
              "air                     1.0   3126.0\n",
              "J                       1.0   3095.0\n",
              "=                       1.0   3053.0\n",
              "enforcement             1.0   3021.0\n",
              "host                    1.0   3021.0\n",
              "First                   1.0   3009.0\n",
              "hit                     1.0   2872.0\n",
              "parliament              1.0   2805.0\n",
              "City                    1.0   2797.0\n",
              ".-                      1.0   2793.0\n",
              "Bernie                  1.0   2745.0\n",
              "hand                    1.0   2738.0\n",
              "O                       1.0   2730.0\n",
              "watch                   1.0   2718.0\n",
              "record                  1.0   2670.0\n",
              "independence            1.0   2473.0\n",
              "West                    1.0   2427.0\n",
              "reality                 1.0   2424.0\n",
              "fear                    1.0   2386.0\n",
              "read                    1.0   2329.0\n",
              "Century                 1.0   2302.0\n",
              "corruption              1.0   2263.0\n",
              "Johnson                 1.0   2253.0\n",
              "popular                 1.0   2234.0\n",
              "room                    1.0   2230.0\n",
              "view                    1.0   2213.0\n",
              "status                  1.0   2184.0\n",
              "R                       1.0   2179.0\n",
              "form                    1.0   2157.0\n",
              "People                  1.0   2121.0\n",
              "Brexit                  1.0   2117.0\n",
              "opinion                 1.0   2100.0\n",
              "base                    1.0   2099.0\n",
              "share                   1.0   2080.0\n",
              "Wire                    1.0   2071.0\n",
              "numbers                 1.0   2056.0\n",
              "terms                   1.0   2044.0\n",
              "coverage                1.0   2016.0\n",
              "operations              1.0   2012.0\n",
              "Foundation              1.0   2008.0\n",
              "Of                      1.0   2003.0\n",
              "Just                    1.0   2001.0\n",
              "Joe                     1.0   1997.0\n",
              "Kelly                   1.0   1989.0\n",
              "cover                   1.0   1973.0\n",
              "poor                    1.0   1966.0\n",
              "favor                   1.0   1944.0\n",
              "body                    1.0   1916.0\n",
              "standing                1.0   1914.0\n",
              "agree                   1.0   1903.0\n",
              "Paris                   1.0   1885.0\n",
              "test                    1.0   1866.0\n",
              "individual              1.0   1824.0\n",
              "screen                  1.0   1819.0\n",
              "Sen                     1.0   1804.0\n",
              "referendum              1.0   1791.0"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "y7PftjyRdI3t"
      },
      "source": [
        "OK! This leaves us with some interesting words to investigate:\n",
        "\n",
        "- `twitter`\n",
        "- `defense`\n",
        "- `gun`\n",
        "- `Brexit`\n",
        "- `popular`\n",
        "- `People`\n",
        "- `corruption`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "xxB7j4YUdI3v",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a991885e-6cd8-4a06-f337-3073ef0ab426"
      },
      "source": [
        "# print neighbors for all high-difference words\n",
        "high_diff_words = ['twitter', 'defense', 'gun', 'Brexit', 'popular', 'People', 'corruption']\n",
        "N_neighbors = 10\n",
        "for word_i in high_diff_words:\n",
        "    print(f'testing word = {word_i}')\n",
        "    print(f'\\tfake news neighbors')\n",
        "    print(fake_news_word2vec_embed_model.most_similar(word_i, topn=N_neighbors))\n",
        "    print(f'\\treal news neighbors')\n",
        "    print(real_news_word2vec_embed_model.most_similar(word_i, topn=N_neighbors))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "testing word = twitter\n",
            "\tfake news neighbors\n",
            "[('www', 0.6070612668991089), ('pic', 0.5979985594749451), ('john', 0.5946311354637146), ('hillaryclinton', 0.5928731560707092), ('??????????', 0.579643964767456), ('Neiman', 0.5753360986709595), ('en_US', 0.5644693374633789), ('sdk', 0.5569521188735962), ('WomensMarch', 0.5483077168464661), ('ConcernedStudent1950', 0.5445438623428345)]\n",
            "\treal news neighbors\n",
            "[('microblog', 0.7688511610031128), ('crotch', 0.7561038136482239), ('Instagram', 0.7461926341056824), ('Weibo', 0.7108451128005981), ('Wheels', 0.6860888004302979), ('LinkedIn', 0.6857026219367981), ('dwelling', 0.6821672916412354), ('Meals', 0.6682783365249634), ('morals', 0.6675081253051758), ('writings', 0.6665583252906799)]\n",
            "testing word = defense\n",
            "\tfake news neighbors\n",
            "[('role', 0.6575046181678772), ('position', 0.615018367767334), ('stance', 0.5982798337936401), ('servant', 0.5973563194274902), ('chief', 0.5921223759651184), ('rejection', 0.5802915096282959), ('alliance', 0.5768030285835266), ('favor', 0.5766576528549194), ('edict', 0.5753903985023499), ('charge', 0.5717446208000183)]\n",
            "\treal news neighbors\n",
            "[('defence', 0.8125608563423157), ('foreign', 0.7119519114494324), ('commerce', 0.6920163631439209), ('transportation', 0.6221017241477966), ('development', 0.6210939884185791), ('interior', 0.6208276748657227), ('management', 0.6205630302429199), ('finance', 0.6163516640663147), ('energy', 0.6149525046348572), ('infrastructure', 0.5878212451934814)]\n",
            "testing word = gun\n",
            "\tfake news neighbors\n",
            "[('firearm', 0.7084372639656067), ('guns', 0.675535261631012), ('bathroom', 0.6285750865936279), ('concealed', 0.6272762417793274), ('firearms', 0.6060128808021545), ('background', 0.5869278907775879), ('corrosion', 0.5847285389900208), ('weapon', 0.5572037100791931), ('license', 0.5546659231185913), ('suit', 0.5484351515769958)]\n",
            "\treal news neighbors\n",
            "[('abortion', 0.7308399081230164), ('environmental', 0.6454980373382568), ('religious', 0.6391414999961853), ('stricter', 0.6206373572349548), ('labor', 0.6195254921913147), ('tighter', 0.6161639094352722), ('strict', 0.6066237092018127), ('traffic', 0.591049075126648), ('gay', 0.5809123516082764), ('excessive', 0.5805349349975586)]\n",
            "testing word = Brexit\n",
            "\tfake news neighbors\n",
            "[('Rouhani', 0.5822790861129761), ('2020', 0.5712632536888123), ('Duterte', 0.5619748830795288), ('gloom', 0.5583631992340088), ('militarism', 0.5552449226379395), ('victory', 0.5518925189971924), ('Candidate', 0.5406746864318848), ('vigilantism', 0.5378959774971008), ('BREXIT', 0.5374472141265869), ('Paris', 0.5348734855651855)]\n",
            "\treal news neighbors\n",
            "[('divorce', 0.752288281917572), ('Britain', 0.71241694688797), ('EU', 0.6959599256515503), ('bloc', 0.6788729429244995), ('Merkel', 0.603806734085083), ('negotiations', 0.6034669280052185), ('Brussels', 0.5970825552940369), ('exit', 0.5765312910079956), ('future', 0.5760590434074402), ('Barnier', 0.5666030645370483)]\n",
            "testing word = popular\n",
            "\tfake news neighbors\n",
            "[('unpopular', 0.7170221209526062), ('margin', 0.612238883972168), ('narrow', 0.6082300543785095), ('favorable', 0.5576472282409668), ('polls', 0.5438762903213501), ('subtle', 0.5399017930030823), ('slim', 0.536993145942688), ('progressive', 0.5352452397346497), ('politician', 0.533599317073822), ('demographic', 0.5272344946861267)]\n",
            "\treal news neighbors\n",
            "[('Latino', 0.6789015531539917), ('nationally', 0.6518588066101074), ('leaning', 0.6468833088874817), ('Hispanic', 0.6320254802703857), ('turnout', 0.6268044710159302), ('winning', 0.6250858902931213), ('battleground', 0.6168834567070007), ('favorite', 0.6163577437400818), ('blacks', 0.6119670271873474), ('conservative', 0.6118586659431458)]\n",
            "testing word = People\n",
            "\tfake news neighbors\n",
            "[('folks', 0.7258380651473999), ('idiots', 0.7249968647956848), ('patriots', 0.7130868434906006), ('Dems', 0.6993215084075928), ('Conservatives', 0.6629174947738647), ('progressives', 0.6459220051765442), ('bikers', 0.6314448714256287), ('guys', 0.6221922636032104), ('liberals', 0.6141989231109619), ('racists', 0.6126310229301453)]\n",
            "\treal news neighbors\n",
            "[('Liberty', 0.6743597984313965), ('Country', 0.6648508906364441), ('Resistance', 0.6522029638290405), ('Women', 0.6285005807876587), ('Movement', 0.6239476203918457), ('Black', 0.6174302101135254), ('Holy', 0.610730767250061), ('Cities', 0.6094937324523926), ('Church', 0.6083824634552002), ('Enquirer', 0.6021800637245178)]\n",
            "testing word = corruption\n",
            "\tfake news neighbors\n",
            "[('hacking', 0.7536702752113342), ('secrecy', 0.7455185055732727), ('ongoing', 0.7365700006484985), ('investigations', 0.7021944522857666), ('litigation', 0.6990571022033691), ('deception', 0.6867972016334534), ('collusion', 0.6756966710090637), ('mounting', 0.6746975183486938), ('violations', 0.6740468144416809), ('theft', 0.6686791181564331)]\n",
            "\treal news neighbors\n",
            "[('graft', 0.9236050248146057), ('criminal', 0.7297093272209167), ('bribery', 0.7296901345252991), ('scandal', 0.7112309336662292), ('scandals', 0.7107619047164917), ('abuse', 0.6732050180435181), ('rebellion', 0.6699951887130737), ('fraud', 0.6699525117874146), ('alleged', 0.6678439378738403), ('lawsuits', 0.6600227355957031)]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:7: DeprecationWarning: Call to deprecated `most_similar` (Method will be removed in 4.0.0, use self.wv.most_similar() instead).\n",
            "  import sys\n",
            "/usr/local/lib/python3.6/dist-packages/gensim/matutils.py:737: FutureWarning: Conversion of the second argument of issubdtype from `int` to `np.signedinteger` is deprecated. In future, it will be treated as `np.int64 == np.dtype(int).type`.\n",
            "  if np.issubdtype(vec.dtype, np.int):\n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:9: DeprecationWarning: Call to deprecated `most_similar` (Method will be removed in 4.0.0, use self.wv.most_similar() instead).\n",
            "  if __name__ == '__main__':\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-iSG1OjHdI3x"
      },
      "source": [
        "This reveals consistent connotation differences in the way that fake news articles frame common issues.\n",
        "\n",
        "- `twitter` is associated with activity on the platform (`WomensMarch`) in fake news, and associated with other social media platforms in real news\n",
        "- `defense` is associated with more general aspects of defending, and associated with government departments in real news\n",
        "- `gun` is associated with different components of owning guns (`firearm`, `concealed`) in fake news, and associated with controversial issues and restrictions (`abortion`, `strict`) in real news\n",
        "- `Brexit` is associated with conflict in fake news (`victory`), and associated with European politics in real news\n",
        "- `popular` is associated with political polling in fake news (`margin`, `politician`), and associated with election outcomes for sub-populations (`Hispanic`, `blacks`) in real news\n",
        "- `People` is associated with \"national\" identity (`patriots`, `liberals`) in fake news, and associated with more diverse identity (`Women`, `Black`) in real news\n",
        "- `corruption` is associated with investigations into hacking (related to Clinton's emails?) in fake news, and associated with financial crime in real news"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xfVfRqsVdI3x"
      },
      "source": [
        "### Visualizing word embeddings\n",
        "Similar to the topic models, we may find that we can explore associations between words more effectively by visualizing the word embeddings in 2-dimensional space."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "U2giS9IidI3x"
      },
      "source": [
        "Let's use Principal Components Analysis again to convert the word embeddings to 2-dimensional space."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KFQWZ_EidI3z"
      },
      "source": [
        "from sklearn.decomposition import PCA\n",
        "from sklearn.preprocessing import StandardScaler\n",
        "def shrink_embeddings(embeddings):\n",
        "    num_viz_components = 2\n",
        "    pca_model = PCA(n_components=num_viz_components)\n",
        "    mini_embeddings = pca_model.fit_transform(embeddings)\n",
        "    # rescale to N(0,1)\n",
        "    scaler = StandardScaler()\n",
        "    mini_embeddings = scaler.fit_transform(mini_embeddings)\n",
        "    return mini_embeddings\n",
        "real_news_word2vec_mini_embeds = shrink_embeddings(real_news_word2vec_embed_model.wv.vectors)\n",
        "fake_news_word2vec_mini_embeds = shrink_embeddings(fake_news_word2vec_embed_model.wv.vectors)\n",
        "# get word to index lookup\n",
        "real_news_word2vec_vocab_idx = {word : index for index, word in enumerate(real_news_word2vec_embed_model.wv.index2word)}\n",
        "fake_news_word2vec_vocab_idx = {word : index for index, word in enumerate(fake_news_word2vec_embed_model.wv.index2word)}"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nwJ4-_jedI30"
      },
      "source": [
        "Let's just try to plot the top-K words to keep things from getting too messy."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "jlvqQvjkdI31",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "af912a4c-2365-4cda-90df-c827bb4bcf02"
      },
      "source": [
        "top_k_plot_word_count = 100\n",
        "top_k_plot_words = word2vec_vocab_news_word_counts.sort_values(inplace=False, ascending=False).iloc[:top_k_plot_word_count].index.tolist()\n",
        "print(top_k_plot_words[:50])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['.', ',', 's', '-', 'Trump', 'said', 'The', ':', ')', 'I', '(', 'U', 't', 'S', 'will', 'people', 'President', 'one', 'also', 'It', 'Reuters', 'Clinton', 'Donald', 'Obama', '?', 'government', 'can', 'He', 'Republican', 'House', 'In', 'year', '/', 'told', 'United', 'just', 'We', 'like', 'state', 'campaign', 'time', 'A', 'election', 'two', 'president', 'States', 'last', 'But', '$', 'country']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JmdnIkvydI32"
      },
      "source": [
        "## code to plot each word as an annotated point\n",
        "import matplotlib\n",
        "%matplotlib inline\n",
        "import matplotlib.pyplot as plt\n",
        "def plot_embeddings(embeddings, embedding_vocab, vocab_to_idx):\n",
        "    embedding_vocab_idx = list(map(lambda x: vocab_to_idx[x], embedding_vocab))\n",
        "    height, width = (10, 6)\n",
        "    plot_fig = plt.figure(figsize=(height, width))\n",
        "    ax = plot_fig.add_subplot(1,1,1)\n",
        "    # plot invisible points to make words easier to see\n",
        "    ax.scatter(embeddings[embedding_vocab_idx, 0], embeddings[embedding_vocab_idx, 1], color=(0.,0.,0.,0.))\n",
        "    for i, (word_i, index_i) in enumerate(zip(embedding_vocab, embedding_vocab_idx)):\n",
        "        position_i = embeddings[index_i, :]\n",
        "        # make words semi-transparent for easier browsing\n",
        "        ax.annotate(word_i, position_i, color=(0.,0.,0.,0.8))\n",
        "    plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "Ij0bKD4IdI33",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        },
        "outputId": "7b3f0e97-5df1-490b-99fc-a05beaf1a384"
      },
      "source": [
        "# plot real news embeddings\n",
        "plot_embeddings(real_news_word2vec_mini_embeds, top_k_plot_words, real_news_word2vec_vocab_idx)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4EA26Wz_dI35"
      },
      "source": [
        "In real news articles, the politicians tend to cluster together (`Obama`, `Clinton`, `Donald`) while the more America-related words form a different cluster (`American`, `United`)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VAvrQ9LMdI35",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        },
        "outputId": "07b88c1c-d553-4206-9156-0c74e62d1c3d"
      },
      "source": [
        "# plot real news embeddings\n",
        "plot_embeddings(fake_news_word2vec_mini_embeds, top_k_plot_words, fake_news_word2vec_vocab_idx)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlsAAAFlCAYAAADcXS0xAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdd1xXZf/48dcBBGQJJg5wB5qIC0fYfWOZRmribYosR67QXLklJ7lTy0GpqDhYIo5K0pzpT81UwJniIBBUHJCgIijr/P4gzlcUV4GYvp+Ph4+bz+dc5zrXOdbNu+t6n/elqKqKEEIIIYQoGTqlPQAhhBBCiFeZBFtCCCGEECVIgi0hhBBCiBIkwZYQQgghRAmSYEsIIYQQogRJsCWEEEIIUYL0SnsAT1KhQgW1Zs2apT0MIYQQQoinio6OTlFV1fLh71/qYKtmzZpERUWV9jCEEEIIIZ5KUZSEor6XZUQhhBBCiBIkwZYQQgghRAmSYEsIIYQQogQ9V7ClKMpKRVFuKIry+wPflVcUZaeiKBf++l+Lx5z7yV9tLiiK8sk/HbgQQgghxL/B885srQbaPfSdD7BbVVVbYPdfnwtRFKU8MAV4G2gBTHlcUCaEEEII8Sp5rmBLVdV9wM2Hvv4fsOavn9cAnYs49UNgp6qqN1VVTQV28mjQJoQQQgjxyimOnK1Kqqpe/evna0ClItpYA5ce+Hz5r+8eoSiKt6IoUYqiRCUnJxfD8IR4NklJSURERJT2MIQQQrxiijVBXlVVFVD/YR/LVFVtpqpqM0vLR+qCCVEiNmzYwLBhw1iyZAne3t78+eefpT0kIYQQr4jiKGp6XVGUKqqqXlUUpQpwo4g2V4D3HvhcFdhbDNcW4h/LyMjA398fPz8/Lly4QNOmTSlbtmxpD0sIIcQrojhmtjYDBW8XfgL8WESb7YCzoigWfyXGO//1nRClTlEUFEXh9u3bAFhZWWFkZFTKoxJCCPGqeK6ZLUVR1pI/Q1VBUZTL5L9hOBsIVxSlH5AAuP3VthkwUFXV/qqq3lQUZRoQ+VdXU1VVfTjRXohSUbZsWSZOnMi3335LSkoKf/zxBwMHDsTQ0LC0hyaEEOIVoOSnWb2cmjVrpsreiOJFSUpKYv/+/Zw5c4Zq1arRv3//0h6SEEKIfxFFUaJVVW328Pcv9UbUQrwIGRkZ3Lp1CwAjIyNq1aqlLSkKIYQQ/5QEW+K1l5OTw8yZM7l16xZpaWlUrlyZ6dOnl/awhBBCvCIk2BKvPTMzM/z8/EhKSiI6OhoXF5fSHpIQQohXiGxELcRfTE1NqVu3bmkPQwghxCtGgi3xrxMWFoarqysTJ04kKyuLQYMG4eXlxY4dO5g2bRpxcXGPPXffvn2sXr26yGOmpqbUqVOHiIgI5syZU0KjF0II8bqRZUTxr7N+/XqWLFlCxYoVOXXqFAChoaEAODs7P/HcVq1a0apVqxIfoxBCCFFAgi3xUgsJCeHHH/Pr5Hbu3JmLFy9y5coVhg4dSocOHfj+++9JTU3Fy8uLOXPmMHXqVIYPH46dnR0HDx7ku+++Iy8vD3Nzc5YsWUJERAQxMTGMHTuWffv2ERAQQHZ2Nubm5kyfPp3y5csXuv6uXbtYtmwZurq6mJiYsHz58tJ4DEIIIf7FJNgSL62YmBg2b97MmjVrUFWV3r17M23aNH777Tf8/f0xNzfH3t6eoKAgFixYUOjc1NRUpk+fzooVK7CysiqylEPjxo1ZvXo1iqLwww8/sGbNGkaMGFGozfLly/n222+pWLEid+7cKdH7FUII8WqSYEu8tI4fP07r1q21fQpbt27NsWPHnuncU6dO4eDggJWVFZD/xuHDbty4wRdffEFKSgrZ2dlYW1s/0qZRo0b4+vrywQcf8P777/+DuxFCCPG6kgR58dqaM2cO7u7urFu3jgkTJnD//v1H2owfP55BgwZx/fp1evTooRU/FUIIIZ6VBFvipdWkSRP27t3LvXv3yMzMZM+ePTRp0uSZzm3QoAFHjx4lKSkJoMhlxPT0dCwtLQH46aefiuzn8uXL2NvbM3DgQCwsLLh+/frfvBshhBCvK1lGFC+tt956CxcXF3r16gXkJ8g/ax0sCwsLJk6cyOjRo1FVFQsLCxYvXlyozYABAxg3bhxmZmY0b96cK1euPNLPwoULSUxMRFVVWrRoga2t7T+/MSGEEK8V2YhaCCGEEKIYPG4jallGFEIIIYQoQRJsCSGEEEKUIAm2hBBCCCFKkARbQgghhBAlSIItIYQQQogSJMGWEEIIIUQJkmBLCCGEEKIESbAlhBBCCFGCJNgSQgghhChBEmwJIYQQQpQgCbaEEEIIIUqQBFtCCCGEECVIgi0hhBBCiBIkwZYQQgghRAmSYEsIIYQQogRJsCWEEEIIUYIk2BJCCCGEKEESbAkhhBBClCAJtoQQQgghStA/DrYURamrKMrxB/7cVhRl+ENt3lMU5dYDbSb/0+sK8W+3cuVK3Nzc8PDwwMvLi99//53Q0FDu3bv31HOftZ0QQojSp/dPO1BV9RzQGEBRFF3gCvB9EU33q6ra8Z9eT4hXwcmTJ9m/fz/BwcHo6+uTlpZGdnY2a9eupUOHDhgaGj7x/GdtJ4QQovQV9zJiG+APVVUTirlfUUJkJqV0pKSkYG5ujr6+PgDm5ubs3r2b5ORkBgwYwIABAwCYNWsWPXv2xM3NDX9/fwDCwsIeaXfo0CH69OlD9+7dGTduHBkZGaVzY0IIIR6hqKpafJ0pykrgqKqq3z70/XvARuAykASMVlX19GP68Aa8AapXr940IUHitpLk4uJCUFAQ5ubmxdJOPJuMjAz69+/PvXv3aNGiBc7Ozjg4ODzynG/fvo2ZmRl5eXkMHDiQMWPGYGtrW6hdWloaY8aMYdGiRZQtW5Y1a9aQlZXFp59+Wsp3KYQQrxdFUaJVVW328Pf/eBnxgQvoA52AL4o4fBSooapquqIoHYAfANui+lFVdRmwDKBZs2bFFwkKMjMz8fHx4caNG+Tm5tK2bVtthsTc3Bx/f39mzZrFmTNnuH//Pm3atGHAgAGFZlIK2h06dAh/f3+ysrKoWrUqU6ZMwcjIqLRv8V/DyMiI4OBgjh07RlRUFD4+PgwdOvSRdjt37mTTpk3k5uaSkpJCfHw8traF/9U5deoUcXFx9OvXD4Ds7GwaNGjwQu5DCCHE0xVbsAW0J39W6/rDB1RVvf3Az1sVRVmsKEoFVVVTivH64il+++03LC0tWbhwIQDp6elERETg7++vzaQMHjy40EzKhQsX8PDwICQkRGuXlpZGQEAAixcv1mZSQkJCZCblOeno6NC0aVOaNm2KjY0NP/30U6HjSUlJBAUFERgYiJmZGb6+vty/f/+RflRV5e2332bmzJkvauhCCCGeQ3HmbHkCa4s6oChKZUVRlL9+bvHXdf8sxmuLZ2BjY8Phw4dZtGgRx44dw8TE5JE2O3fupHv37nh5eREXF0d8fPwjbR6cSfHy8uKnn37i6tWrL+IWXhkJCQkkJiZqn8+dO0eVKlUwMjLS8q3S09MpW7YsJiYm3Lx5k4MHD2rtH2zXoEEDTpw4waVLl4D8GcwH+xZCCFG6imVmS1EUY+ADYMAD3w0EUFV1KeAKfKYoSg6QCXioxZksJp5J9erVCQkJ4cCBAyxZsoTmzZsXOi4zKS9ORkYGc+fO5c6dO+jq6lKtWjUmTJjA9u3bGTJkCJaWlvj7+1O3bl1cXV2pVKkSjRo10s7v0qVLoXa+vr5MmDCBrKwsAAYNGkT16tVL6/aEEEI8oFgT5Itbs2bN1KioqNIexisjOTmZcuXKoa+vz/79+/nhhx+4fPky8+fPx8rKivPnzzNlyhRCQkJIS0vDw8ODoUOH4uLigru7u9YuNTWVHj16sHTpUqpVq0ZmZibJycnyy10IIcRrrcQT5MXLLzY2loULF6Kjo4Oenh4+Pj6cOnVKZlKEEEKIEiQzW0IIIYQQxeBxM1uyN6J4JUkRViGEEC8LCbZEsUlKSmLbtm1/+/zc3NxiG8vatWsl2BJCCPFSkJytV1Rubi66urov9JpXr15lw4YNLF++nHr16nH27Flq167N1KlTiY+PZ/78+WRkZGBubo6vry8VKlTA29ubunXrcvz4cT788EMcHByYN28emZmZlClThqVLl2JoaIifnx/R0dFkZWXh5uZGly5diI6O1mp//fHHH9SrV49p06axbt26R4qwCiGEEKVFgq0XbMWKFWzduhULCwsqVapEvXr1tDIK9+7do2rVqkyePJmbN28yefJkAgMDgfxZoxEjRrBu3TpiYmKeKXDZt28f9vb2REVFkZ6ezqRJk2jSpAkRERHs3buXzMxMLl26RM+ePcnOzmbLli3o6+uzaNEizMzMuHz5Ml999RWpqakYGhoyceJEatasia+vL8bGxsTExJCSksLnn39OmzZt8PPz4+zZs1y4cIFmzZqxYcMGpk6dSnh4OHv37uXrr7/GwsKCHTt2sHjxYiZPngzkVzwPCgoiOzubrl27Mnv2bOzs7Lh79y4GBgb88MMPmJiYEBgYSFZWFv369cPR0RHIr0+1fv16KlSoQL9+/Thx4sQjRViFEEKI0iTLiC/QmTNn+OWXXwgLC8PPz4+YmBgAJk+ezLBhwwgLC8PGxobly5dTs2ZNcnJySEpKAvKLjTo7O5OTk8PcuXP56quvCA4OplOnTixevFi7RkHg0qNHDyB/hiswMJBRo0axfPlyrd0ff/zBvHnzCAwM5LvvvsPQ0JDQ0FAaNmyoVTKfMWMGY8aMITg4mOHDhzN79mzt/JSUFFasWMGCBQvw8/MDYOjQodjb2+Po6MgXX+Tv2tS+fXsOHTpEbGwsgwcPxsvLi4CAAK5f/7+NBpydnYH8Qp8VKlTAzs4OAGNjY3R1dTl06BBbtmzBy8uL3r17c+vWLa1oZ/369alYsSI6OjrUqVNHe15CCCHEy0Jmtl6g48eP8+6776Kvr4++vj5OTk5kZmZy584dHBwcAOjYsSPjxo0DoG3btuzYsYPevXuzY8cOZs2aRUJCgha4QH4wVaFCBe0aBYFLgffffx+AevXqFQpEmjVrhpGREUZGRpiYmODk5ATkV5m/cOECGRkZnDhxAh8fH+2cgjIPAO+99x46OjrUrl2bmzdvFrrmX5sFaIyMjHjzzTdZtWpVkc/F0NDwqc9uzJgxtGzZstB30dHR6Ovra591dXWLNe9LCCGEKA4SbL3EnJ2dGTduHO+//z6KolC9enViY2OfK3ApU6YMkL8P34OByINBio6OjvZZURRyc3NRVRVTU1NCQ0OLvM6D5z9cPuTatWucPHmShg0bsm3bNho0aMD333+vfZeTk0NiYiK1a9cudF6NGjVISUnhzJkz2NnZkZGRgYGBAY6OjmzYsIHmzZujp6dHYmIilpaWT3x2BdvZyDKiEEKI0ibLiC9Q48aN2bdvH1lZWWRkZHDgwAHKli2LmZkZx44dA2Dr1q00bdoUgKpVq6Krq8uKFSu0GasaNWqQmprKyZMnAcjJySEuLq7Yx2psbIy1tTW7du0C8gOq8+fPP/EcIyMjMjMzqVGjBuvXr8fV1ZU7d+7g7u7OnDlz8PPzw9PTEy8vL06cOPHI+WXKlGH27NnMmTMHT09PBg0axP379+ncuTO1a9eme/fuuLm5MWPGjKfOYBUUYR0wYMAT2wkhhBAlTWa2XiA7OzveffddPDw8KF++PDY2NpiYmPDll19qCfLW1tZMmTJFO+eDDz5g4cKFbN68GcgPSObMmcPcuXNJT08nNzcXT0/PR2aJisP06dOZNWsWAQEB5OTk4OzsTJ06dR7b3tbWFh0dHSIjI+natSvTpk3TjtWpU6dQzliBZcuWFfpsZ2fH6tWrH2k3ePBgbem0QNOmTbXAFGDs2LHaz+7u7ri7uz/1Hp8kKSmJ4cOHEx4eXmi8ZcuWpWfPnv+obyGEEK8PqSD/gmVkZGBkZMS9e/f49NNPmTBhAm+99VZpD6vYFBWg/FtJsCWEEOJ5yN6IL4kZM2YQHx/P/fv36dix4ysVaAFYWVm90EArLy8PHR1ZDRdCCPHykmDrBZsxY0ZpD6HULF26FDMzM7y8vABYvHgxFhYWZGdns2vXLrKysmjdurWWZzVq1CiuX79OVlYWHh4edOnSBQAnJye6dOnCkSNHGDduHPv372ffvn3o6uri6OjI8OHDS+0ehRBCiIfJlIB4YTp16sSWLVuA/Bmp7du388Ybb3Dp0iXWrFlDaGgoMTExHD16FIApU6YQHBxMUFAQYWFh3Lp1C4DMzEzs7e1Zu3YttWrVYs+ePYSHhxMWFka/fv2KbbwPl7B42vdCCCFEUWRmS7wwVlZWlCtXjnPnzvHnn39St25dzpw5w6FDh+jevTuQn9N26dIlHBwcCAsLY8+ePQBcv36dxMREGjRogI6ODm3atAHAxMQEAwMDpk2bxn//+1+tXlhxKFeuHHfu3Cn03a1bt7Cysiq2awghhHj1SbAlXqjOnTsTERHBn3/+yf/+9z+OHDlCnz59tCXCAtHR0Rw+fJhVq1ZhaGiIt7e3VlTVwMBAy9PS1dVlzZo1REZGsnv3bsLDw1m6dGmxjNXIyIgKFSoQGRlJ8+bNuX37Nr/99huenp7F0r8QQojXgywjiheqdevWHDx4kNOnT9OyZUtatmzJjz/+SEZGBgA3btzg5s2bpKenY2ZmhqGhIRcvXuT3338vsr+MjAzS09P5z3/+w8iRI59aC+x5ffnll6xYsQIvLy8GDhzIp59+StWqVYut/7y8vGLrSwghxMtJZrbEC1WmTBmaNWuGqakpOjo6ODo6Eh8fT58+fYD82aRp06bRsmVLNm7ciKurKzVq1MDe3r7I/jIyMhg5ciRZWVmoqsrIkSOLdby1a9fG398fyE/wT01N1Y4VV4J/48aNi3XMQgghXjKqqr60f5o2baqKV0tubq7q6empJiQklMr1P/30U/X06dOqqqrq0KFD1du3bz/zuVeuXFG9vLxUVc2/j06dOqnbt29Xp0+frubl5am5ubnq559/rkZHR6uqqqq3bt1SVVVV7927p3br1k1NS0tTVVVVmzZtqu7YsaM4b0sIIcRLAIhSi4hnZGbrNeLk5MT+/ftL7fpxcXEMHz6c1q1bU7169VIbR4FFixY9V/uSSPAXQgjx6pNgS7wwtWvX1rYdKi5JSUkMHTqUevXqcfbsWWrXrs3UqVM5efIkCxYsIDc3l/r16+Pj41No82wAFxcXgoKCMDc3Z8uWLQQFBaEoCra2tkydOpXU1FRmzZrFtWvXgPxlweJO8BdCCPHqk//Hfw1lZGTw2Wef0b17d9zd3fl//+//ARAYGEhYWBgA33zzDQMHDgQgMjKSiRMnltp4nyYhIYFu3bqxYcMGTExMCAkJwdfXl9mzZ7Nu3TpycnLYsGHDY8+Pi4sjICCApUuXsnbtWkaPHg3AvHnz8PLyIjAwkDlz5jBt2rRiT/AXQgjx6pOZrdeQgYEB8+bNw9jYmLS0NHr37k2rVq1o0qQJwcHBeHh4cObMGbKyssjJyeHYsWM4ODi8sPE9756ElSpVokyZMsydO5f27dsza9YsypYtqy1VduzYkfXr12uV6x8WGRlJ27Zt6dmzpzbTBXDkyBHi4+O1dnfv3iU7O7tYE/yFEEK8+iTYeg2pqsp3333H0aNH0dHR0WZj6tWrR0xMDHfv3kVfX5+33nqLM2fOcPz4cW2252WkKAp2dnbY2dkRGRlJenq6tmT3T+Tl5bF69epCy495eXn8/vvvzJ49W/vO09OzyNpbj8sJK828OSGEEC+eBFuvoZ9//pnU1FSCg4PR09PDxcWF+/fvo6enh7W1NRERETRs2BBbW1uioqK4dOkStWrVKu1hA+Dt7Y29vT1RUVGkp6czcOBArl27RlhYGIcOHeLcuXMcPnyY+/fvU7VqVebPn09UVBRnz56lV69e/Pbbb7Rr1w47Ozuys7MZPXo0SUlJXLx4kWrVqgFw+/ZtzMzMcHR0JCwsjF69egGwbt065s+fj7u7+0uR4C+EEOLfQXK2XkPp6elYWFigp6dHVFQUV69e1Y41adKEoKAgHBwcaNKkCRs3bqRu3bov1X6Aubm5BAYGMmrUKEJCQqhRowa7du1i06ZNXL58malTp+Lj44OjoyMhISH89ttvTJ48mcDAQN5++22twnzB24Fbt27lk08+ITIykn79+vHNN98AMGbMGGJiYvDw8KBbt27s27cPR0dHRowYUZq3L4QQ4l9GZrZeQ+3bt2fEiBG4u7tjZ2dHzZo1tWONGzcmICCABg0aULZsWQwMDGjSpMkLHd/TNoB+//33AahXrx7Xrl2jXLlyfPbZZ6SmplK9enXKli1L+fLl8fX1BeCdd97B09OTnJwc9PX1qVatGhkZGWRlZXHr1i169epFeno6b775JgEBAWRkZNC/f38yMzMBGD9+PA0bNqR3797Ex8fj5eVFx44d8fDwwM/Pj+joaLKysnBzc3vkrUQhhBBCgq3XSEGukLm5OatWrSqyTYsWLTh8+LD2edOmTS9kbA962gbQZcqUAUBHR6fQdjeVK1fm+vXr7N27F1tbW+rXr0/Dhg25ePEix44do1KlSuzYsYNDhw5hZGQE5OdgBQYG8uuvv+Lm5gZA+fLlWbx4Mfr6+iQmJjJhwgSCgoIYOnQoQUFBLFiwAMh/NiYmJgQGBpKVlUW/fv1wdHSUjaqFEEIUIsGWeOk8aQPoh+t0GRoaEh4eTnR0NGXKlCE4OJgZM2Zw9uxZfHx86N+/P3p6enTu3JlatWqRm5uLnl7+P/ZvvPGG1s+dO3e4e/cuADk5OcyZM4dz586hq6tLQkJCkeM8dOgQsbGx7N69G8hfnk1MTJRgSwghRCESbImX0pdffslXX33F/PnzAZ55A2gdHR0++eQTxo4dS05ODuvWrcPJyYnq1asTHx9PXl4e9evXB/KXIc+dO4ebmxu2trYYGBgAEBoaSvny5Vm7di2qqtKyZcvHXm/MmDFPPC6EEEJIsCVeSg9uAP2gZcuWaT+bm5sTEREBQNOmTalQoQKJiYlUr16dsLAwFi9ezO3btzl06BCenp40bNiQnJwcEhMTAdDX12fs2LHY2dmRlpbGyZMnMTc3Jz09nYoVK6Kjo8PmzZu1pUojIyOteCmAo6MjGzZsoHnz5ujp6ZGYmIilpSVly5YtyUcjhBDiX6bYgi1FUS4Cd4BcIEdV1WYPHVeAhUAHIAPorarq0eK6vhAZGRnMnTuXO3fuoKurS7Vq1ZgwYQJdunRh7ty5pKenk5ubi6enJ7Vr135sP66urowdO5YtW7bQsmVLLXiytbVFR0cHT09PXFxc8PDw4OrVq3Tv3h1VVbGwsODrr79+UbcrhBDiX0LJ36S6GDrKD7aaqaqa8pjjHYCh5AdbbwMLVVV9+0l9NmvWTI2KiiqW8QkhhBBClCRFUaIfnmyCF1tn639AoJrvEGCuKEqVF3h9IYQQQogXrjiDLRXYoShKtKIo3kUctwYuPfD58l/fCSGEEEK8soozQf6/qqpeURSlIrBTUZSzqqrue95O/grUvAHZEkWIf2jlypVs27YNHR0ddHR0GD9+fIlvit2iRQtsbGzIycnB2tqaqVOnYmpq+tz9JCUlcfLkSdq1a1cCoxRCiBen2Ga2VFW98tf/3gC+B1o81OQKUO2Bz1X/+u7hfpapqtpMVdVmlpaWxTU8IV47J0+eZP/+/QQHB2tvZ1aqVKnEr2tgYEBoaCjh4eGYmZmxfv36v9XP1atX2bZt23Odk5ub+7euJYQQJalYgi1FUYwVRTEt+BlwBn5/qNlmoJeSzxG4parqVcRLLSIigjlz5jzXOcOGDXukAvzDli5dypEjR4D8gqIP/0Leu3cvcXFxzzdYIDo6mpMnT2qfN27cyJYtW567n1dBSkoK5ubm6OvrA/mlMl70f8A0bNiQGzduAHD58mWGDh1Kjx496N+/PxcvXgTA19dXKwwL4OTkBICfnx/Hjh3Dy8uL0NBQ8vLyWLhwIb169cLDw0Pb3SA6Opr+/fszcuRIunXrRmZmJp9//jmenp64ubmxY8eOF3rPQgjxsOJaRqwEfP/X3nV6QKiqqtsURRkIoKrqUmAr+W8ixpJf+qFPMV1bvGQWLVr01DYDBw7Ufi4Itrp166Z9t3fvXpycnIos0ZCbm4uurm6R/UZHR1O2bFkaNmwIQNeuXZ93+K8MR0dHVqxYQZcuXWjRogXOzs44ODi8sOvn5eVx5MgR/ve//wEwY8YMvvjiC6pXr87vv//O7NmztU3Bi/Ks2yMBnD17lvDwcKysrPjll1+wtLRk4cKFQH5lfyGEKE3FEmypqhoHNCri+6UP/KwCg4vjeqL4jBo1iuvXr5OVlYWHhwddunRh8+bNrF69GhMTE+rUqaPNjPj6+mJgYMC5c+e4efMmU6ZMYcuWLZw8eRJ7e3tt42cXFxeCgoLIyMhg2LBhNG7cmJMnT2Jpack333yDgYEBvr6+ODk50aZNG3r37k10dDQ1atSgXr16uLu7s2/fPoKCgjA2Nmbr1q3069cPY2NjjI2NMTEx4ezZs8THx2NlZUXjxo2ZPn069+7dY8OGDeTl5TF79mwsLS3JzMykXLlyrFixggEDBlCpUiVu3LjB3bt3iYiIwMzMDG9vb+zt7YmKiiI9PZ1JkyZhY2PDtm3bCgWA/zZGRkYEBwdz7NgxoqKi8PHxYejQobi4uJTode/fv4+Xlxc3btygVq1aODo6kpGRwYkTJ/Dx8dHaZWVlPVe/j9seqUyZMtSvX1/bJsnGxob58+ezaNEinJycXvhG6kII8bAXWfpBvISmTJlCcHAwQUFBhIWFcePGDfz9/QkICCAgIOCRpbw7d+6watUqRo0axYgRI/Dy8iI8PJzY2FjOnz//SP+JiYm4ubkRHh6OqalpoeUiyN9gWk9Pj/fff5+EhAQGDhzI7du3adWqFbVr18bBwYGqVauSkpJCpUqVGD9+PDdv3iQsLIzTp09TtWpV6tevz5o1a7CysqJr165kZWUxdepU9u7dS79+/Wjbti1//vknp0+fZtiwYWzduhVPT0+WL6JpZwsAACAASURBVF+ujSM3N5fAwEBGjRrF8uXLi1za/DfS0dGhadOmDBgwgHHjxvHLL7+U+DULcrZ++uknVFUlPDwcVVUxNTUlNDRU+7NhwwYAdHV1tSr9eXl5ZGdnP7bvMWPGaOdv3rxZm9l6sGp/9erVCQkJwcbGhiVLlhT6exZCiNIgwdZrLiwsDE9PT3r37s3169fZunUrTZs2xcLCgjJlyuDs7FyovZOTE4qiYGNjwxtvvIGNjQ06OjrUrl2bpKSkR/q3tramTp06QP5ehFevFk7TMzExQV9fn5iYGH755ReaN2/OsWPHSE1NpUqVKpQvX56UlBRu3rxJjx49OH78OK1bt+bOnTuMHTuWP/74gzVr1mhBYUJCAjo6OoWWD6tUqYKJiQnZ2dk4ODgQHR1NZGQkR48eZfTo0ezYsYMDBw7QqVMnYmJiSEpK4ttvv+Xy5ct4eXmxcOFCVFVl4cKFWFpa4u7uruUBzZ8/n7fffpuxY8fStWtXJk6cSHEVCv6nEhIStK2JAPr27UtsbKz2OSgoqND2R39XWFgYrq6uTJw4sdD3hoaGjBkzhuDgYAwNDbG2tmbXrl0AqKqqBedWVlbExMQAsG/fPnJycoDHb49UcDwxMZHMzMxHxpOcnIyhoSEdOnSgZ8+enD179h/foxBC/BOyN+JrLDo6msOHD7Nq1SoMDQ3x9vamZs2aT0xML1hSVBSFMmXKaN/r6OgU+SbY09ro6uqyYMEC+vbty4EDBwgPD+fOnTvcvn2bJk2aULNmTXbu3Imenh4WFhZa8DBnzhx69OiBvb09qampWrJ1cnKytpz0LOrUqYOBgQGTJ0+mRo0auLi4UK5cOYYMGUJsbCyhoaEA/PLLL5w7d4633nqLJUuW0LNnTy3/KTk5mdGjR1OhQgX69evHiRMnaNy48TOPoaQ8vH1RTk4OOTk5pKWlYW5uXmzXWb9+PUuWLKFixYqPHKtbty62trZs376d6dOnM2PGDAICAsjJycHZ2Zk6derw8ccfM3LkSDw9PZ97e6SiXt6IjY1l4cKF6OjooKenV2jpUgghSoMEW6+x9PR0zMzMMDQ05OLFi/z+++/cv3+fo0ePcuvWLYyNjdm1a5c2M1USNm7cyKpVqzh37hwtWrTgyJEj5ObmEh8fT2pqqvYLOjc3l2+//ZYbN25w8OBB6tatS0REBGFhYVhYWNCiRX6lkZ07d2Jubo67uzsGBgbY2dlhbW3NqVOnuH37Nu3atcPU1JTr16/TuXNnTpw4QXZ2NmXKlCE9PZ2EhASysrJYvXo1e/bsAfKD0gkTJlClShViYmL45ptvcHBw4MyZMwBUqlSJihUrkpSUxJkzZ/j888+xs7NjypQpVK5cuVB+GuTPDu7fv5+UlBS++OIL7t69S05ODl988QVNmjTh0KFD+Pv7k5WVRdWqVZkyZQpGRkbP/Wzr1avHypUrtc9OTk64ubkRGhrKoEGDCrVNTU1l1qxZXLt2DcjP5WvUqBHu7u6sWLECExMT2rRpw6hRo/joo4+YPHkyH330Ebt37+bKlStaLtixY8ewtramd+/eTJgwAVtbW+rVq8ehQ4fYsGEDlStXplGjRly5coXDhw/z448/MnLkSBwcHPj111+5cOGC9twvXLhAXl4eurq6HDx4EGdnZwYPHsyJEyeoW7cux48fJyIigh49etC0aVPtXlq2bEnLli2f+3kJIURJkWXE11jLli3Jzc3F1dUVPz8/7O3tqVChAt7e3vTp04e+fftSq1YtIL845vr165k+fTpeXl6cPXuWxMRE7t2799TrhIaGFtkuKSmJlStXYmJiQrly5QgODqZhw4Z8+eWX2Nvbc//+fYYMGUJycjLGxsZcu3aNsLAwRo4cydGjR1m9ejU1atSgQoUKWpBgaWnJlStX0NXVpVmzZuzcuROA+vXr4+TkROXKlYmNjSUuLo7+/fsD+bN0APPmzaNq1ao0aNDgkRIJN27coHPnztSqVUvLGZo6dSobN27U3oycO3cudnZ2jB49mvbt2zN37twnPpdt27bRsmVLQkNDCQsLo27duqSlpREQEMDixYsJCQnBzs6OkJCQpz7jZ+Xm5sbPP//8yBt68+bNw8vLi8DAQObMmcO0adMAaNSoESdOnCAuLo6qVaty7NgxAE6dOkXDhg0ZP348lpaW+Pv7k5SURN26dQkLC2Pw4MFMmTJF6z8uLo4lS5Ywc+ZMIL8MxNKlS/nmm2+YNGkSzZo1Y926dRgYGHDgwAFycnKYO3cuX331FcHBwXTq1InFixdr/WVnZxMUFESPHj2K7dkIIURJkZmt15i+vn6RZRqaNm1Kp06dtM8nT55k/vz5HDt2DH19fdLS0sjOzsbS0pJ79+5haGiovYkI+bW5IL+uU3h4OC4uLnTo0IEPP/yQ2bNn8/HHH5OXl0dsbCzt2rWjZs2aODg4MHbsWGJjY5k/fz7VqlUjOzsba2tr/Pz8WLZsGXp6eujp6TFy5Eg2btzIwYMHURSFpUuXYmZmBkDlypWpWbMmrq6udOrUiR9++AFHR0fOnDlDbGws5cqVw8LCAlVVtXOaN2+OnZ0dJ0+exNraWisZMG7cOO2e7O3tOXLkCIaGhvTo0YNt27axbt06AgMD2bhxo/ac2rZtC0CHDh20fh7Hzs6OqVOnkpOTw3vvvUedOnWIjo4mLi6Ofv36AflBRYMGDR7bh7e3N8OHD8fOzu6J1ypgbGzMRx99RFhYGAYGBtr3R44cIT4+Xvt89+5dMjIyaNKkCUePHqVKlSq4urqyadMmbty4gampaaGkdIDjx49rAWbz5s1JS0vj7t27ALRq1arQ9f7zn/+gp6eHjY0NeXl52kyUjY0NSUlJJCQkEBsby+DB+S8w5+bmUqFCBe38h3MJhRDiZSbBlniqoopjhoWFkZyczIABAzA3N8ff359Zs2Zx5swZ7t+/T5s2bRgwYEChdseOHWPevHm4ubmxdOlS9uzZQ2RkJDNmzADyi1jOnz+fWrVq4eLigpOTE/7+/to4CvK/CnJxCmakHs4Fmzx5MiEhIaxatYrTp0/z7bffkpycTLVq1Vi3bh3BwcF89dVXT7zncuXKYWxsjJubG9WrV6dmzZrUrl2bdevWcfv2bbp27cobb7zxTM/vcW/bOTg4sHz5cg4cOICvry/du3fH1NSUt99+W5sBKgleXl507969UAmIvLw8Vq9erf0dF3BwcGD9+vVcu3aNwYMHs2fPHnbv3v3c5RQeDsye9nepqipvvvkmq1atKrI/Q0PD57q+EEKUJllGFE/l6OjI9evX6dKlC7Nnz+bo0aN4eHhoy0cFAdHgwYO1EhLR0dFcuHBBa+ft7c27775Lq1atCAgIYMmSJWzZsoWkpCS2bNlCQkIC8+bNIysri1atWtGvXz9++uknDh06RI8ePVi2bBmRkZHamE6ePMnChQu1HKTLly/j7e1NZGQkoaGhzJ49Gzc3NywtLUlJSeHgwYN8/vnnAPzxxx/aTEq7du20RPcGDRrw6aefYmVlxfbt26lZsybh4eF4enqiKAqff/45b731Ft27d9eCDRsbG20WsFGjRjRq1AgXFxd+/vlnrc2Db9tNmDBBq6ju6+vLxIkT+fjjj7G3t2f+/Pncu3ePwMBAPv74Y8aNG8eff/5JYmIiMTExeHt706NHD4YMGUJKSkqhv6O8vDx8fX0LLbU9jpmZGR988AE//vhjob/jsLAw7XPBm4KVKlUiLS2NS5cuYW1tTePGjQkKCiqyOGqTJk34+eefgfw8N3Nzc4yNjZ86nqLUqFGD1NRUbTeAnJycv7WjgBBCvAwk2BJPVVAcc8KECVhYWODj46MtFT5o586ddO/eHS8vL+Li4gotS128eJF69epx6tQpbZls5syZGBoasnPnTjZu3IixsTFt2rRh0aJFeHh4YG5ujoODA8HBwfTp04fDhw9z69YtID+4aN68OeHh4ejr67N7924WL16MnZ0dO3bswMPDg23btvHjjz8SEhLCnDlzGDJkCD169Hjsm3ijRo0iJCQEDw8PLl++jImJyXM9pzFjxrB582Y8PDzYunUro0ePBuDjjz/m6NGjeHp6oqqqVszz4MGDbN++XWvfsWNH1q9fz9q1azE1NWXv3r20a9eO2NjYJ+Yv5ebmMnHiRKpXr/5I4vvj9OjRg7S0tEJjj4mJwcPDg27dumk1sCB/CbVgU/gmTZqQnJxc5NuWAwYM0Prw8/Pjyy+/fK7n96AyZcowZ84c/Pz88PT0xMvLixMnTvzt/oQQojQpL0tNoKI0a9ZMjYqKKu1hiIfs3r2bn376idjYWIKCgjA3NycpKYlBgwYRGBiImZkZvr6+NG3aFBcXF1xcXOjcuTO3bt2iWbNmbNu2TVsm8/LyolOnTpw5c4aJEycSGRnJ7NmzycjIYPfu3Sxbtkx7O62g/lWDBg1o2bJloZwtfX19+vbtS8eOHUlLS+PAgQOkp6czZ84czp07h66uLgkJCfz666+Pva979+5hYGCAoijs2LGDbdu28c033zzSrkWLFtjY2JCTk4Ouri4fffQRXl5e6Og8/b9dcnJy6NKlC2vXrmXMmDHUrl0bZ2dnli5dSqtWrVi+fLm2WXRBvlb37t3p06cPVatWBf4vf+m7777D29ubO3fu8MEHH9C3b9/HXjc0NJQuXbpoy2/Dhg1jxowZmJqaPnXMQgghno2iKNGqqjZ7+HvJ2RJPlZCQgKIo2uzGuXPnqFKlCklJSWRkZGBubk56ejply5bFxMSEmzdvcvDgQe11fCMjIypXrsyhQ4fo06cPX331FZcuXaJ8+fJcuXKF1NRUcnJySE9P5z//+Q9t2rRh+fLlRdYBK5gVejjPpyAHSFEULX8rNDSU8uXLs3btWlRVfWo5gJiYGObMmcPvv/+OnZ0d8+bNK7JdQYV0gJs3bzJx4kTu3r3LgAEDnvos9fT0sLa2JiIigoYNG2Jra0tUVBSXLl3CysqqyHyt2NjYJ+YvNWzYkMjISLy8vB6by7R27Vo6dOigHX+W/SuFEEIUDwm2xFM9XByzWrVqTJgwge3btzNkyBAtd6tu3bq4urpSqVIlGjX6v60yu3TpQkBAAKdOneLgwYP4+voyfvx4jh8/jp6eHmlpaRw6dIjBgweTm5vLoUOH8PHxKbIO2NNERETg5OQE5NcRq1ixIjo6OmzevFlLUn+cJk2asHbt2ud6NuXLl2fChAn07NkTb29vVFXFz8+P6OhosrKycHNzo02bNri7uxMbG0t2djapqamEhoZSoUIFsrKyyM3NxcXFhQYNGvDll1/So0cPcnJyMDU1ZcCAATRq1Ijr16/zySefkJWVhaqq9OrVi8aNG7Nz507atWvH0aNHGTZsGDVq1ODs2bOPfUmh4GWGgv0rzc3N2bJlC0FBQSiKgq2tLVOnTn2uZyCEEOLJJNgST/VwccwC7u7uuLu7a58fLP9QVLvr168ze/ZsLl68SF5eHj179mT48OFs376dlJQU0tPTuXHjBhMnTtRmsTZu3Iirqys1atTA3t7+ucbt6urK2LFj2bJlS6HK5A/KzMzEx8eHGzdukJubS//+/dmwYYNWTsHJyQlPT0/279+PgYGBtqx48+ZNZs6cyZUrV4D8PSNTU1P5+uuv+eWXX6hVqxZ2dnZs2rQJR0dHZsyYwahRo+jUqROpqan4+fnRp08fTp48SXx8POnp6VhYWGjFUStWrMjZs2eZO3cu69ato1q1apw/f57KlSuTm5ur5Vulp6fz4YcfsmTJEvz9/Tl//jxr1qwBYODAgdpLCiEhIfj7+z+SrxYXF0dAQAArV67E3Nyc27dvP9czFkII8XQSbIkXplKlSsyfP/+R7wvyuh72uDpgAPv379d+9vb2LvJY9erVC71hN2zYsEf6+e2337C0tNRqYqWnpxdKDs/MzMTe3p5BgwaxaNEivv/+eyC/gKmDgwPz5s0jLy+Pd999l4SEBLZv3061atUA2LRpEwYGBiQmJlKmTBns7OwwMTHB0tKSbt268fbbb3P8+HGaNm3Knj17GDx4MNeuXWPnzp1cu3aNWbNmsXHjRvr378/Zs2f55ZdfCpVmSEpKomXLlnTu3BnIT1DfuHEjPXv2JDc3l5SUFOLj47G1tS3yGQJERkbStm1bLQgrqD0mhBCi+MjbiOKJXFxcCr219qyio6O11/Yhf1ueLVu2FOfQnsnTxm9jY8Phw4dZtGgRx44de+QNxDJlymjLkm+99Za22XZkZCSurq4AXL16lTJlynD+/Hn+/PNPrZBn7dq16devH46OjgCFAqWCvSUTExNxdnamffv2mJqakpqaCqDNhNWvX5/hw4drez8+7MEcraSkJIKCgliyZAlhYWH897//5f79+8/zuIQQQpQACbZEiYiOji70qn7Xrl356KOPSnFERatevTohISHY2NiwZMkSli9fXuj4g4n4urq6j2yknZqaysyZM7Xl1MzMTBo0aEBgYCCbNm2iXbt2ZGZmPvb61tbW1KhRg3PnzvHmm29SsWJFNm/ezIkTJ+jTpw9JSUnMnDkTExMTbcYtLy/vke12gCJfUihgZGRERkbGI+c0b96cXbt2aSU1ZBlRCCGKnywjCs3WrVsJCwsjJycHe3t7fHx8nnpcR0eHgwcP8t1335GXl4e5uTmTJk1iw4YN6Orq8vPPPzNmzBgiIyMpW7YsPXv25Pz588ycOZN79+5RtWpVJk+ejJmZGd7e3tjb2xMVFUV6ejqTJk16rkrlo0aN4vr162RlZeHh4UGXLl20Y0XlZjk7O7N9+3YtH83ExITTp08/9Tr379/n5s2btGrVilq1atGuXTs+/vhjUlJSyMrKolKlSnTv3p2srCxMTExYsmTJY/sqeIvy3Llz5Obm8sEHH7Blyxbi4+Pp3Lkz69ato3Llyty8eZPp06fz448/oquri4+PT6HtawDq1KnzxJcUHnyZoUDt2rXp27cvn376Kbq6utStW/exuXdCCCH+Hgm2BADx8fHs3LmTlStXoqenx+zZs7Vq4E86/s477zB9+nRWrFiBlZUVt2/fxszMDFdXVy24AgpVf588eTJjx47FwcGBpUuXsnz5ckaNGgXk15AKDAzk119/Zfny5c9UEb3AlClTMDMz4/79+/Ts2ZM2bdpox4rKzcrKysLX1xdLS0tMTEyIjIykR48exMbGPvE6R44c0YKfK1eusH37dpo0aULDhg2xsrLi+PHj5OXlcfr0aWxtbenXrx+DBg1iwYIFfPvtt2zcuJHy5cvz9ddfc+vWLS5cuEDFihVZtWoVBgYG7Nu3j759+9K1a1cqV66MqqqkpKQUWfMrPDy80OenvaRQ4MGitB07dqRjx45Pfb5CCCH+Hgm2BJAfDMXExNCrVy+2b9+OkZERlSpV4ubNm9y5c6fQccgvAFq+fHlOnTqFg4MDVlZWwNMTrCdNmsT169e17V46duxYaMPn999/H8h/A7IgP+pZhYWFaQVQr1+/TmJionbMxsaG+fPns2jRIpycnGjSpAnnz5/H3t5eWzo8cuQI69evZ9myZdp5Q4cOxcvLC4CFCxdqAVz58uWLDH4sLCwIDQ0lNzeXe/fuYWxsTFpaGr1796ZVq1bExcVhamrK2rVrSUpKYsiQIdrbgh999JFWGHX69OnMmjWLgIAAcnJycHZ2pk6dOs/1PIQQQrwcJNgSAKiqSseOHRkyZAhOTk7aG3116tThxx9/xNLSUjv+oH379j3XdcaNG8fRo0cfe/zBDYofzo96kicVQIX/y806cOAAS5YsoXnz5rz77rtP7dfNzQ03N7dnHkcBVVX57rvvOHr0KDo6Oty4cYObN28yfPhwBg8erAV9mzZt0s4p2KII8vdT9PPze+7rCiGEePlIgrwA8hOld+/ezc2bN4H8ROmrV69iZmZGcnIyzZs3Z968eVrSdWJiIh988AENGjTgwIEDuLq64uXlRdeuXUlMTERPT49Vq1bh6emJm5sbZ86cAWDkyJGoqsqxY8eYNWsWrq6unD59ulAe0d/xtAKoycnJGBoa0qFDB3r27MnZs2epWbMmSUlJXLp0CcjPSStqg+W/4+effyY1NZXg4GBCQ0N54403uH///lMT8oUQQrx6ZGZLAPmJ0p999hlDhgzh7NmzDBo0iDFjxpCWlsY777xD7dq1sbOz0/bTy8vLIysrCwsLC+rXr8/p06epUqUKFSpUoFKlSpQtW5bk5GQqV67MF198Uagu1pAhQ1i4cCF37tzhP//5D5MmTWLs2LHaG3F/R8uWLZ9YADU2NpaFCxeio6ODnp4ePj4+6OvrM2XKFMaNG0dubi7169ena9euf3sMDyooUqqnp0dUVBRXr14F8oO+cuXK0aFDB0xNTfnhhx+A/3tb8HGbZAshhPj3ko2oxSMKNlq+ceMGtWrVwt/fHx0dHby9vbXK6mlpafTs2ZOIiAi2bdvGypUr+eijj2jdujXVq1cnMTGRwYMH88EHH2g5UkChPjZu3MimTZu0Apxjx47F2dm5lO/+nylYgk1LS2PEiBFkZGRgZ2fHqVOnWLRoEQkJCY8EfXZ2dqxbt45169Y98ragEEKIfw/ZiFo8s4KNlu/du8eQIUMIDw/Hw8MDPT09CoLzB/Oh2rVrh729PQcOHODzzz9n/PjxNG/e/JEcqU8//VQ7p6AAZ2BgIGZmZvj6+r4SBTgLZvDMzc2L3DjaysqqyA2xH35bUAghxKtDcrbEYxkaGjJmzBiCg4PJzc2lSpUqxMTEALBr1y6t3ZUrV7C2tsbDw4N3332XCxcukJycTPv27QvlSEVERGiFTp9UgFMIIYR4lcjMlniiunXrYmtry/bt2+nZsyc+Pj5s2rSJ//73v1qbnTt3snXrVvT09HjjjTfo27cvp0+f5vz583h5eWnLZX/88Yd2zpMKcAohhBCvEsnZEiXmwRISkF9IMyYmhrFjx5KamsqsWbO4du0akF/9XQIuIYQQ/2aSsyVeuPv372sFQQFu3bql1baaN28eXl5eNG7cmGvXrjFkyBBt7z8hhBDiVSLBligxBYn2BQpmtiC/Wnt8fLx27O7du2RkZGBkZPTCxymEEEKUJAm2RKnIy8tj9erV6Ovrl/ZQhBBCiBIlbyOKUuHo6EhYWJj2+fz58yV+zbCwMFxdXWndujWrV69+5vOSkpLYtm1byQ1MCCHEK02CLVEqxowZQ0xMDB4eHnTr1u2F5GutX7+exYsXs2fPHnr37v3I8cftxXj16lUJtoQQQvxt8jaieC3MnDmTzZs3U6NGDf73v/9x+fJlxo4di6+vL/r6+pw7d45GjRrx3nvvMW/ePO28FStWMGjQIOLj47G2tqZjx46Fkv4f1rdvX1auXPnc44uOjiYoKIgFCxY88znLli2jbNmy9OzZk6VLl+Lg4ECLFi2e+9pCCCGKR4m9jagoSjUgEKgEqMAyVVUXPtTmPeBHoCAjepOqqlP/6bX/DUJDQ+nSpQuGhoalPZTX2vjx4/ntt9/w9/cvVI4C4MaNG6xatQodHR1GjBjBuHHjaNSoERkZGRgYGDB06NBnDoT+TqBVHAYOHFgq1xVCCPF0xbGMmAOMUlXVDnAEBiuKYldEu/2qqjb+689rEWgBrF27lnv37hV5LC8v7wWPRhSlbdu26Ojk/6vQqFEjvvnmG8LCwkhPT0dXV/e5+nJycgLyZ6q8vb0ZO3YsXbt2ZeLEidpWR2fOnKFv3754enrSq1cvMjIyCvWxbNkygoKCtM9ubm4kJSUB+cFcly5d6NevHxcvXtTa+Pr6snv3bgBcXFzw9/ene/fuuLu7a+1SU1MZNGgQbm5uTJs2jY4dO5KWlvZc9yeEEOL5/eNgS1XVq6qqHv3r5ztADGD9T/t9kbZs2YKHhweenp5MnjyZpKQkBg4ciIeHB5999plWePPBX2jw9F+sYWFhJCcnM2DAAAYMGKCdM3/+fDw9PQkICGDUqFFaf4cPH2b06NHPPf4WLVrg5eWFm5sbI0aM4M6dO//kcRSp4F4f9uAzmTZtGnFxcX/7Gi/iPtq3b//Id2XLltXuo3fv3lhYWHDp0iX69u1bKKB5XufOnWP06NGsX7+eK1eucOLECbKzs/Hx8WH06NGsXbuWJUuWYGBg8Ez9xcTEsH37dkJDQ1m0aBFnzpx5bFtzc3NCQkJwdXXVArfly5fTvHlzwsPDadOmjfbPtRBCiJJVrAnyiqLUBJoAh4s43FJRlBOKovysKEr9J/ThrShKlKIoUcnJycU5vCLFxcUREBDA0qVLWbt2LaNHj2bu3Ll07NiRsLAw2rdvz9y5c5/aT1G/WD08PLC0tMTf3x9/f38AMjMzsbe3Z+3atfTv35+LFy+SmpoKwObNm+nUqdNz30NBPavw8HDMzMxYv379c/dRHCZNmkTt2rX/9vkvw31cvnyZBQsWMGbMGOzs7Lh48SJGRkaPzD49i/r161OxYkV0dHSoU6cOSUlJJCQkUKFCBezs8id/jY2Nn3n27NixY7Ru3RpDQ0OMjY1p1arVY9u2bt0agHr16nH16lUAjh8/zocffgjAO++8g5mZ2XPfkxBCiOdXbMGWoigmwEZguKqqtx86fBSooapqI8AP+OFx/aiqukxV1WaqqjaztLQsruE9VmRkJG3btsXc/P+zd+9xOd7/A8df1905lYocasthJTlUEsMcctwcYjMqCZtTRizGnPYlOZszm5jDiEQOG3PIYZLDd1QkRDIRCxWdj6rr90ffrp/IaSKHz/Px8Nh9X/d1+HzutfX2ud7X+20IgIGBAZGRkXz22WcAdOnShYiIiGeep7RfrKVRqVS0b98eAEmS6Nq1K/v27SM9PZ3z58/zySefvNR8bGxsSEhIAIoCh5EjR+Lu7q4EdlC0GjVr1iz69etHz549lRym3bt3M2/ePOVcXl5ehIeHK+8XLlyIs7Mz33zzjRIgPmzohjpqIQAAIABJREFU0KHKasvJkyfp27cvffr04ZtvvgHg4sWLfP3117i5uTFw4EBu3LihXHfcuHH8/ffffPHFFyxduhQbGxvu3LmDt7c3Tk5OWFpa8sknn5TJPH777Tc2btzIN998Q3Z2dok5+Pv7U7t2bbp06YK6ujqSJOHj48Off/6JpaUl/v7+T51HbGwsI0eOZOzYsSVWntTU1J74tOOj1NTUePjBlby8vOc67mHF9ctUKhX5+fkvfLwgCIJQdsok2JIkSYOiQGuTLMs7Hv1cluU0WZYz/vd6L6AhSVLlsrj266SmpqbkWRUWFvLgwQPls4eLcz7tF6uWlpaSHwRF+TV79+4lKCiIDh06vHCO0MMKCws5ffq0suIxc+ZMxo0bx8aNG/Hy8mLOnDnKvrdv32b9+vUsXryY2bNnP/MXenZ2NtbW1mzduhV7e3t++eUXAK5fv84333xT4nZbcnIyM2bM4Mcff2Tz5s3MnTsXgJo1a7J69Wr8/f0ZNmwYP/30k3JMdHQ0NWvWZMuWLQQFBXHkyBFq1qzJ5cuXSUpK4vDhw+zfv/+l5rF7924ePHhAr169iIyMxN7enkqVKinBL8D3339Phw4dmD9/PuPGjWPu3LksWLCAq1evEh4ejpub21PnkZ2dzezZs5k9ezY3btzg7t27JcZQo0YNkpKSlEAsKyvrsZ+V6tWrc/nyZQAuX76sBO729vYEBweTm5tLVlbWY4n+z2Jra8vBgwcB+Ouvv0hLe/TvRIIgCMKrUBZPI0rAGuCSLMsLn7BPNeCuLMuyJElNKQry7r3stctCkyZNGDt2LH379qVixYqkpaVha2vLgQMH6NKlC/v27aNRo0YAmJqacunSJTp27EhISMhzrRgU34IqXjl7lImJCSYmJqxZs4aff/75X82huAdhQkICtWrVolmzZmRlZXHu3DkmTJig7PdwINKxY0dUKhXm5uaYmZk9MzdJpVLRqVMnoGi1b9y4cQBcvXoVCwsLgoKClH3Pnz+Pvb09pqamAMrtqoyMDKZOnUpcXBySJJX4/po2bcquXbv46quviIyMREtLi08//RR/f3/i4+MZNGgQlStXRpIkYmNj2bNnz2PzMDQ0pEePHgwfPvyF5wFw+PDhEiuL/2Ye+vr66OnpoaGhQcWKFbl9+zZVq1ZVPtfQ0GDOnDnMmzeP3NxctLS0Hvv33r59e/bs2YOzszMNGjTA3NwcgLp169KpUyf69OmDkZGRcivyeQ0ZMoTJkyezZ88ebGxsqFSpkmiPJAiC8BqURbueT4B+wHlJkorvt00CzAFkWfYFegHfSJKUD2QDrvIbUuCrdu3aDBw4kCFDhqCmpoaVlRXjxo1j2rRpbNiwASMjI6ZOnQrAF198wZgxY+jTpw/NmzdHR0fnmefv2bMnnp6eSu5WaTp37kxycjK1atX6V3MoznXKycnB09OTrVu34uTkhL6+fonehM/y8ModPPv2VVZWFnfu3GH48OFs27aNypUrc/HiRTZs2EBCQgKRkZF07NgRCwsLNm/eTHh4OMOHD2f+/PlERUXRvXt3+vfvz82bN7G1tUVLSwtHR0diYmIICwtj0KBBeHl50bdvX2xsbNDV1UVdXZ2CggL8/PzQ0tLC3t6e2bNnExUVxfHjx9HX11fm4eTkRLdu3di7dy9RUVGsWrXqqfM5cuQI33//famfbd++HW1tbbp27cqKFStwcHBg/vz5nDlzhm7duin7FQd6jRs3xtHRUVm1evi89erVUyrY+/v7o1KpaNy4MY0bN2bUqFHMnDmzxGrZwwYOHMjOnTtZsGBBiQDe29tbeb179+4S1yqet56eHsuXL0dNTY3IyEiioqJEuyRBEITX4KWDLVmWjwPSM/ZZDix/2Wu9Kt26dSvxCxPA19f3sf2MjY1LtHkZNWoUgPKLstjDv1hdXFxwcXFR3pd26yciIoIvvvjiX4+/mLa2NuPGjeO7776jd+/emJmZcejQITp06IAsy8TExFCnTh0ADh06RLdu3YiPj+eff/6hZs2aZGVlsW3bNgoLC0lMTOTChQvKuQsLCzl8+DCdOnVi//792NnZcfToUT788EOqVq2KoaGhUkbg/v376Ovrs3TpUgYOHEhSUhLdu3fn4MGD/PLLLwwaNAhHR0cMDAwoKCjAwcGBHTt2kJGRwZo1a7h27RoLFy7E39+f3bt3k52dzYkTJ/jnn3/49ttvadCgATVr1iQqKoqtW7eSnZ1Nfn4+FSpUICsriwcPHhAdHc21a9dYt24dycnJfPzxx/j5+ZGeno6TkxP169fnzz//xNDQEFmWuXDhAikpKXh4eBAZGYmXlxcNGzZkzpw5xMfH8+WXXyq33TIyMqhSpYryPb6MzZs306VLF6UO29KlS1/qfE9z584dJkyYgCzLaGho8MMPP7yyawmCIAj/TzSiLmfu7u7o6Ojg5eVVJuezsrLC0tKSoKAgZsyYwezZs1mzZg35+fl06tRJCbaqVavGgAEDyMzMZOLEiWhqamJra4upqSm9e/emVq1a1K1bVzmvjo4OFy5cYPXq1RgbGzN79mymTZvGRx99BECnTp1YsqSolq2trS3u7u5MmjSJa9eukZqaSu/evalXrx4jR46kbdu2SlAkyzInTpzg3r17WFlZMWjQIBYvXkxMTAyVKlVi/fr1mJiYkJCQQEJCAqtWrSI3NxdLS0uaNWtGeHg4SUlJZGVloa6ujoaGBunp6QQEBFBYWIipqSmZmZmcO3eO4OBg7t+/T1xcHPfv3+fDDz8kMjKSrl27kpqair6+PitXrqRLly6MHj0aIyMjLC0tGTt2LDExMRgYGHD48GFatmzJgAEDUKlUJSq2FxYWcvz4cS5cuEBeXh4ZGRlAUWmQlStXYmhoyN9//421tTXTp09ny5YtSmkQQ0NDVq5ciZOTE35+fhgaGvLdd99x9+5d8vLycHV1pWfPni/1s2Fubv5CK52CIAhC2RDBVjnbuHHjS5/j0dWyRYsWKa+XLVtW6jEff/wxkyZNKrFNkiRmzJjxXNdIS0sjNDQUIyMjFi9eTEFBAcbGxtSqVYsTJ07QokULWrRogbu7O3Fxcdja2ioBn6mpKTExMYSHh1OjRg2OHTuGt7c3VatWZfXq1ejp6aGpqYmfnx9dunQhLS2N0aNH880337Bq1SquXLnCrl27uHLlComJiTRu3JjAwECOHTuGu7s7+fn5SJKErq4u5ubmpKSkkJiYiCzLfPDBB7Rv355Zs2Yxbtw4vvrqKxwcHEhMTGTBggWkpqZSrVo1tm/fjiRJpKeno6+vr7TGgaKG1r///jv29vYsWbKEorTFot6KPXr0YNCgQeTl5TFo0CCqV6/O7du3iY6OJjAwkMqVKzNo0CClNMimTZuUQOxRU6dOxcDAgNzcXPr160f79u2pWLHiM34aBEEQhDeNaET9DktPT1dqVSUmJj4xH+lhwcHBJQqTpqam4ubmhpubG506daJz5864ubnRtWtXPvvsM/744w92797N3r17MTMz4+zZs4+dszgYefi9kZER+/btA0CWZSwsLPD392fw4MGMHDmSKVOmoKamxrRp0/jwww85ffo0ULRKtGnTJoyMjEhOTiY7O5vLly/j4eHB0qVLlVIOenp6VK1alc6dOzNz5kwMDQ2RJAlbW1s0NTU5f/48165dY+fOncydO5c//viD3Nxc9PT0iI2NpXfv3vz5559oa2vj6empJOSnp6cTGRnJpUuX8PLyokuXLsq8/vrrLyX5vFevXqSmpiq5V89bGqRYq1atCAgIoE+fPnz11VfcvXuXuLi4Z/77EwRBEN48Ith6R8XHx7Nr1y4l2DIxMVFqT3l7e5cod1CsoKCA4OBgYmNjlW0VK1bE398ff39/evXqhZubG/7+/tSrV4/27duXKFvQrl27Ek8lFktKSiIyMhIoKhVhZ2fHRx99RHR0NK6urixYsIDg4GBu3rwJFCXmx8XFkZWVRVZWFqampnh4eKBSqUhNTeXcuXOcPHkSCwsLDAwM6Ny5M7m5uWRmZlKjRg0AKleuzN27d7l48SJ2dnbk5OSQm5ur3BodOnQod+7coaCggMqVK+Pn54ednR25ubnMnj0bbW1tjh8/jqenp7IyViw9Pb1E2Y9ihw8fZvjw4URGRrJr1y527dpFSEgIO3bs4O+//1b2e56aWxkZGZw6dYqZM2cyYMAArKys/lW9LUEQBKH8iduI75j4+HhGjhyJoaEhO3fuREdHBxcXF1JTU7l8+TJNmzZFS0sLExMTcnJy2LNnD61btyY2NpYHDx4QFxfH6dOnWbNmDaNHj8bd3Z0mTZqgra2NsbExx48f5++//6ZGjRqsXLmSv/76Cz09PXR0dOjXrx+urq4levndv38fWZZxdnZGkiQcHBw4cOAAERERTJ06lebNm7Nq1SpMTExo27YtOTk51KxZE0tLS3R1dfn111/Jy8vju+++w8rKigEDBnDp0iUqVapEo0aNuHv3LtHR0VSvXp3evXuzf/9+OnXqxLJly6hcuTInT57k7NmzSJKEmpoalpaWhIeHY2hoiLGxMba2tly9epXNmzfTunVrhg0bho6ODmlpaYwZM4aOHTvSqVMnbt26RXZ2NlpaWuTm5qJSqcjKymLYsGFER0fzww8/MGnSJPbt20dAQABjxowhMDCQ7OxsfvvtN6U/5t69e9myZQt6enqcP38eHR2dUkuDFBQUYGBgwMmTJ5k+fToZGRlcvXqVFi1avPafKUEQBOHliJWtt8iyZcvYunWr8n7VqlVs2LCBJUuW4OzsjIuLC0ePHuXGjRskJiZiZWVFZmYmxsbGLFq0CCMjI7S1tTl69Cj79u1j/vz5ODo6cvz4cUaPHs2pU6ewsbHBzs4Of39/1q5di4ODA1OmTMHLy4udO3cqNccSEhJYt24dY8aMeeqY4+PjlScHGzdujKWlJevXr2fLli1s27ZN2S8zM5OLFy9y4cIFDAwMsLa2pnLlymzYsIGAgAB8fX0xNTXlo48+IjAwUKmHtXz5cho0aIAsyxw/flxJIq9ZsyZ6enoYGBjQpUsXFi9eTGFhIV27dmXx4sWoq6vj7e3NrVu3+Oeff1i8eDGampoYGBhw/vx5goODcXNzQ09PT1ntO3jwIJcuXVLy3TZs2MDdu3extrZWzlGtWjWOHDmCs7Mzv/76K3FxcTRt2hQLCwtiY2PZsWMHWlpaSJLEzp07MTY2xtPTk8aNG/Pjjz9y7tw5+vTpQ0FBAQUFBYwbN47r16+jqalJ27Zt+eeff965YqTBwcE4ODi8VB9KQRCEN5lY2XqLdOrUiQULFuDs7AzAwYMHGTBgAH/99RcBAQGkpKTQu3dvpTbYihUrqFChArq6uqxevZq4uDjq1KlDpUqVuHfvHikpKWhpaWFmZqY0mjY2Nub+/ftKUVRNTU1cXFyoW7cuiYmJ2NjYcPv2bTp06FCiEv6TVKtWDT09PVQqFbVr16ZJkyZIkoSFhYXSsw+gTZs2aGlpoaWlhYODAxcvXsTR0VH53NTUtESg2a9fP+X1ow8BxMfHo6am9lj+2L17JevoNmnShHXr1jFu3DhatGhBrVq1WL9+PYcOHeLnn39GpVIRERHBTz/9RHx8PN9//z3BwcEYGRkpJS2WL19OTk4OS5Ys4eLFiwwZMoTQ0FDS0tL47bffaNeuHcnJybi7uxMaGsqBAwdo1KgRFSpUICEhgaioKEJCQvD29iYpKYnr169z/fp1mjRpwtKlS2natCnz5s1TbsPOnj2b3r17P7FI7tsoKCgIOzs7goKClIbtgiAI7xKxsvUWsbKy4v79+yQmJnLlyhX09fW5cuUKn332GSqVCmNjY2xsbJSSA8UkSWLr1q1oamqipqZGZmamcssQQF1dvURxy8LCQmRZRl9fnwMHDlCrVi2GDRtGs2bNlIrjDxd0fVovPwMDAyVIUqlUynUerbxeWhL9q1Rcdb9fv35Uq1aNHj16AEWJ7Kamptja2hISEsLZs2eZOnUqP/30EwkJCZw6dYrWrVsTEhJCdHQ0V69eRVdXt0SbpaSkJD7++GO0tbVRU1NT2ifl5OSgpqZGhQoVAKhSpYpSrwvA0dFRCUof/m4ePveL9Fh8G2RlZREREcGUKVNKzfcTBEF4F4iVrbdMhw4dOHz4MPfu3aNTp06lPtV2//59YmJi0NDQ4ObNm3z11Vds2bIFKysrVq9eTadOnWjYsCHNmjVj7dq1JY7V0tIiLy+PChUqYGZmRkhICM2aNWP27Nl8+OGHpY6pevXqHD9+HCjZy+9FHD16lK+//prs7GzCw8MZOXLkC5+j2KOrYKUprrr/qOIg0tbWllu3btGkSRM2b94MFNVES0hIoFevXtSsWZPFixfj7e1N3759n3id4or3UNTbMCsri+XLl2NsbExaWhqZmZnKvu9jNfejR4/SvHlzpd3SpUuXsLa2Lu9hCYIglCmxsvWW6dSpEwcOHODw4cN06NABOzs7Dhw4QGFhIcnJyZw/fx5LS0uOHj3K/v37MTY2VnKDwsLCGD58OObm5mRlZSllEh5mb2/PmTNncHNzY/jw4fz+++8cOXKEc+fOlbo/FPXyS01NxdnZma1btyq9/F6EhYUFHh4efPXVVwwePBgTE5MXPkdZsrCwQFNTk9TUVGWVycTEBB0dHbKzs6lYsSIWFhZ8/vnnStPoYpUrV+b06dPk5uYqOWLbtm2jQYMG2NnZ0atXL1xdXRk+fDhhYWFPHYe2tvZz9eB8WwUFBfHpp58CRT/bYnVLEIR3kVjZesvUrl2bzMxMTExMqFy5Mm3btuX8+fO4uroiSRKDBg1i586dLFu2DE9PT1JTU3FycsLV1ZUVK1YQEhKCLMuYm5tTUFDwWL/AUaNGKW2IABwcHPDz86N79+588803pY5JS0vrib38Hl5herh/36OrT5aWlvj4+Pybr+SVUKlUXLhwgRUrVtC3b19kWcbIyIhNmzYRHBzMhg0bUFdX58CBA/j4+GBqaqoUft26dStr165VGkZ/9dVXSsmJdevWMXv2bJKSksjNzeXOnTulXt/GxgaAzz//nEOHDtGnTx+cnJyeq1ba26K4MO7Vq1eRJImCggIkSeLbb7995beRBUEQXifpDekHXSoHBwf5WX/zF0qKj4/Hy8vrmbfRntfYsWO5desWvr6+rywpu7g6+8NJ70LpcnNzGTlyJL6+vs/1gMKbbMeOHVy+fLlEJ4OhQ4cybNgw7O3ty3FkgiAI/44kSeGyLDs8tl0EW4Lw9ti6dSsFBQX06dOnvIfy0jw8PBgwYECJ2mEBAQHExsYyceLEchyZIAjCv/OkYEvcRhQA2LRpE7///jtQdOvK0dGRUaNGYWdnR2RkJCYmJixcuBAtLS1u3brF3LlzSU5ORltbmx9++IGaNWuW7wTeE/v3739i/8q3zcqVKx/b5urqWg4jEQRBeLVEsPUWKq4S37BhQyIjI6lXrx7du3fH19eX5ORk5Zfx/Pnzyc3NRVtbm6lTp1KjRg12795NSEgIOTk53Lp1i7Zt29KxY0eWLVtGTk4OderUYcyYMRgbG3Pjxg1u3ryJubk5+vr6HD58mC5dujBz5kwmTpyIubk5Fy5cYM6cOfj6+pbzt/Lue/DgAbdu3cLU1LS8hyIIgiC8ABFsvaVu3rzJ3LlzqV27Nv3792f//v2sWbOGkJAQ1q5di4+PD6tXr0ZNTY3Tp0/z008/Kb0Ro6Oj8ff3R1NTk549e6KpqYmbmxuHDx/G39+fAQMGcObMGVq0aMH69evx8vLC2tqa27dvK8VOJ0yYoIxF9Ox7PVJSUtDX1y/vYQiCIAgv6O3OsH2PmZmZYWFh8cTK7BkZGYwfPx5nZ2cWLFhQohFy06ZN0dPTQ1NTk9q1a5OcnIympiZNmjTh2LFjZGRkUFhYiJGREVBU5HT37t2sWLECLy8vKlSogL+/P/PmzaNSpUpoa2szePBgrl+/TlZWFt27dyc/P5+hQ4cSFhZG9+7d8fT0JD09vby+rndCcQ00QRAE4e0igq23lIaGhvK6tMrsK1aswMHBga1bt7Jo0aISv6QfLp6pUqmoU6cOwcHBdO7cmZ07d3LmzBmaNGmi7FPc32/o0KEYGxsDcOjQIWbOnMnYsWOV3olz5sxBV1eXxo0bK0VOT5w4Qdu2bVm+fLlYlXlJBgYGFBQUiIBLEAThLSOCrXdURkaG0grmjz/+eOq+NWvWxMnJiblz57J9+3bu3bvH3bt3OXnyJHPnzkVXV5c//viDFStWYGBgQLt27ZgwYQLr1q2jXr169OjRg1mzZpGQkMCkSZM4duwYY8eOJTc3l+DgYLp3746TkxMpKSkAfPfdd7i7u+Ps7MyOHTte+Xfxb4waNYrExMTyHsZjmjVrRkRERHkPQxAEQXgBIth6R/Xv35/ly5fj5ub2XBXI+/bty9atWxk7diyampp8+umnxMXFcfPmTZKSkjh69CiLFy/m2LFjREZGMmnSJNq0acOdO3ewtrZmzZo1fPnll2hra3Po0CGqVatGWFgYhYWFfPTRRyWuNXXqVDZu3Iifnx8BAQGkpqa+qq/hX1u6dGm5V7EvjbOz8zOD57fdgwcPGDJkyDvVA1IQhPebSJB/Cz1aff1JldkfXjUaPnw4AE5OTjg5OSnbFy9eXOLcFy9epEqVKlStWhWVSkWNGjVITExU8sFSUlJITk4mICCAmJgYevTogY6ODrdv3+bw4cNKlXlXV1e+//57HB0dHxt/QEAAR44cAeDu3bvExcXRsGHDl/tS3hN169bFwcGBwsLCt76o6ZNoaGjQpEkTDhw4QOfOnct7OIIgCC/t3fy/tfDC0tPT6dmzJ1paWhgYGCjbVSqV8ktdkiQKCwuRZZkff/yR48ePU79+fQwMDBg/fjw3b95UjuvcuTMPHjygZcuWJa4THh7OqVOnWLduHZs3b8bKykrkIL2g7t27v7OBVjFHR0f27dtX3sMQBEEoE2Jl6zXavXs3ly5deiP72+nr67Njxw6l3U9x6xx9fX2mTJmi7FejRg3atGnDli1bGDduHMuWLSM6OhorKys2bdrE/v37adKkCXv37kVTU5MKFSqUuE5GRgYGBgZoa2tz/fp1Lly48FrnKbwdLCwsiIqKKu9hCIIglIl3+6/HwisxePBg8vPzcXV1xdnZmRUrVgDQq1cvsrOzadiwIT4+PjRr1uyxY5s3b05BQQG9evVi2bJlaGtrk5ub+7qnILzhVCoVGhoaZGVllfdQBEEQXprojfiQ4srs1tbWXL58mdq1a+Pj40NsbCyLFi0iKysLQ0NDvL29qVy5MleuXGHWrFnk5OTwwQcfMGXKFAwMDBg6dCh16tThzJkz5OfnM3XqVOrXr19iZSs5OZnZs2dz584doOgJPVtb29c21zdBYWEhPXr0wM/PD0NDQ1q1asWxY8eUzx/+vrZv3462tjZdu3bF29ubVq1a0b59e4YOHYqXlxf16tUrx5kIr0L79u0JCgpCXV0swAuC8HYQvRGf040bN5gyZQq2trb4+PiwdetWgoODWbBgAUZGRhw4cICff/6ZKVOmMGXKFL7//nvs7e3x9fXll19+4bvvvgMgJycHf39/zpw5w7Rp00oktENRKx03Nzfs7Oy4c+cOnp6ebNu27bXNMzs7mwkTJpCQkEBBQQGDBw9m2bJlSuATFRXF4sWLWbVqFatWreLWrVvcvHmTlJQU+vfvzxdffEF4eDi+vr7o6upy8+ZNHBwcmDBhAiqViqCgINauXYssy7Rs2ZJRo0YB0KpVK3r27Mnp06dp164diYmJeHh4YGho+NTxfvnll2Uy73c5sfxdkpqaiqGhoQi0BEF4J4j/kz2iatWqygpT586dWbduHVevXmXEiBEAFBQUULlyZTIyMkhPT8fe3h6Abt26MX78eOU8n376KQD29vZkZmY+Vj399OnTxMbGKu8zMzPJyspCV1f3lc6v2H//+19MTExYsmQJUJRLtWzZsifuHxMTw6+//kp2djZubm5K4vvFixcJDAykevXqeHp6cuTIEWxsbFi6dCkbN27EwMCAESNGEBwcjKOjI9nZ2TRo0IDRo0cDsGvXLlauXKmsbD3JqlWr0NHRUXLJSjN79myioqLIzc2lffv2eHh4AEVPYHbs2JFTp07Rvn17Dh8+zKZNm4Cigq0TJ05U3gtvhrCwsMcerhAEQXhbiWDrEZIklXivq6vLRx99xLp160psz8jIeKHzPPq+sLCQX3/9tUQ199fJwsKCRYsWsXTpUlq1akWjRo2eun+bNm3Q0tJCS0sLBwcHLl68iL6+PvXr18fMzAyAzz77jIiICNTV1WncuLHS7qdz586cOXMGR0dHVCoV7du3L/Uaubm5uLm5Ke9TU1Np06bNc89pxIgRGBgYUFhYyLBhw4iJicHS0hIAQ0NDJaA6deoUV65coU6dOuzevZvu3bs/9zWE12P//v2MHDmyvIchCIJQJsT9lEfcuXOHyMhIoOh/+A0bNiQ5OVnZlp+fz7Vr19DT08PAwICzZ88CsHfvXho3bqyc58CBAwBERESgp6eHnp5eies0a9aMgIAA5f2VK1de6bweZW5uzqZNm7CwsGDFihX88ssvqKmpUVhYCDzeXPpJweOzgspHaWlpPfE2npaWFv7+/sqfYcOGkZaWhrOz83PN6eDBg/Tt2xc3NzeuXbumrBympKSUyAX7/PPP2bVrF4WFhRw4cIDPPvvsuc4vvB4PHjzA0dERc3Pz8h6KIAhCmRDB1iNq1KhBYGAgvXr1Ij09HRcXF+bNm8eyZcvo06cPbm5unDt3DoBp06axZMkSXF1diY6OZvDgwcp5tLS0cHNzY9asWSVKJxQbN24cly5dwtXVld69e7/WfC2AxMREtLW16dKlC/369ePy5cuYmppy+fJlAP78888S+x89epS8vDxSU1MJDw9XEtIvXLitAmbcAAAgAElEQVRAfHy8ErjY2dlRv359zpw5Q0pKCoWFhQQFBZUIRB+mq6tbJk+cxcfH4+fnx4oVKwgICKBly5YlnnJ8OPenffv2nDx5kmPHjmFtbU3FihVf+vpC2dHQ0KBr167lPQxBEIQyI24jPkJNTY3p06eX2FanTh1++eWXx/atU6cOv/76a6nn6dy5s5IsX+zh6u2GhobMnj27bAb9L1y9epUlS5agUqlQV1dnwoQJ5ObmMn36dCpUqPBYcGRhYYGHhwcpKSkMHjwYExMT4uLiqF+/PnPnzlUS5ItvFY4cORIPDw8lQf5JtwN79uyJp6fnU1vjFBQU8Ntvv5GUlMT58+fR1NRkz549rF+/nuPHj6Ojo8Pw4cPR0dEhOTkZLy8v9u7dS926dR+7PRoVFcWMGTOoW7cus2fPLjUQFgRBEISyJIKt91Tz5s1p3rz5Y9uf1Bja0tISHx8fAIKDg3FwcGDq1KlUqFDhsZY/UPSAQPFDAg97+HYegIuLCy4uLgBKgnxUVBR79uyhbt26QNETop07d6ZOnTrExsZy9uxZvv76a7p3787QoUPJy8vj7t27WFlZ0apVK+zs7OjXrx/NmzencuXKyrUiIyOZN28eCxcuJDExkYiIiFJrgQmCIAhCWSqTOluSJH0GLAHUgNWyLM955HMtYAPQGLgHuMiyfP1Z533ddbbeJ3v37iUgIID8/HwaNGiAhYUFt2/f5ttvvwVK1rgaOXIkZ8+exdzcnAYNGpCWlkZSUhJBQUFYWlpStWpVtLS0WLhwIcbGxi81roKCAtTU1JT38fHxDBkyhD179gAQGhpKQEAAXbt2Zf369eTk5JCWloaLiwvOzs706tWLvXv3ljhneHg4Pj4+aGlp8dNPP2FiYoKfnx8ZGRlKL0dBEARBeFmvrM6WJElqwE9AR+AWECpJ0i5Zlh/utTEISJZl2UKSJFdgLuDystcW/p3Y2FgOHjzI2rVrUVdXZ86cOejq6nLkyBEl2Dp48CADBw4kNjYWdXV1goODUVdXx8fHh0OHDrFjxw62b9+Ot7c3FSpUYOTIkbi6uqKtrU3Hjh2xsLBg8+bN5ObmsmDBAj744IMnFnItruN169YtqlWrxpdffomfnx+LFy8mOzubmJgYXFxckCSJ1q1bI0kSHh4e2NraolKpqFSp0jP7K1auXJm8vDyio6OZO3cut27dwtfX95V/14IgCIJQFrcRmwJXZVm+BiBJUgDQA3g42OoBeP/v9TZguSRJkvwml69/h4WGhnLp0iX69+8PFBVgNTY2xszMjPPnz2Nubs7169extbUlMDCwxL7Xr1/H1NQUc3NzNDQ0qFy5MtnZ2WRmZtK6dWt8fHzo0aMHn3/+ORs2bGDz5s1s2bKF77777qmFXK9du8aaNWvQ0tIiPDxcGau/vz+ZmZmkpaVhZGTE/Pnz0dfXJy0tjYSEBIyNjQkLC8PKygpdXV2qVKlCcHAwly9fpmHDhkruWXGPx+HDhzNu3LgnJuwLgiAIQlkri2DLDLj50PtbwMdP2keW5XxJklKBSkDSoyeTJGkoMBQQj36/IrIs061bNzw9PUts37VrFwcPHqRmzZo4OjoiSdJj+3p5edGnTx8AjI2NOXDgAK1ataJGjRpoa2ujqanJBx98oORCWVhYUHwr+EmFXAFat26NlpbWY2M9e/YsNjY2ODg4cOnSJSwtLXF2duavv/7ijz/+ID09nSpVqnDv3j0AfHx8mDVrFikpKRw/fpy5c+cq5zI2Nmbx4sWMGjWKKVOm0KBBg7L6SgVBEAThid64BHlZllcBq6AoZ6uch/NOatKkCd999x1ubm4YGxuTlpZGZmYmjo6OrFmzhujoaKW9zsP7qqur89dff3Hx4kW0tbVJSEjg4MGDtGzZEg0NDeX8kiQp71UqFQUFBcDTC7nq6OiUOlZdXV0WL16sBN6rVq0iMzOTxMREgoODmTx5Mvfv3ycsLIwRI0awcOFCfH19lf6JZmZm7Nixg3/++QdXV1eaNWv2WOskQRAEQXiVyiLY+gf48KH3H/xvW2n73JIkSR2oSFGivFAOateuzTfffIOnpyeFhYWoq6szfvx4qlevTq1atbh27Rr169d/bN+4uDhycnL45ZdfOHjwIDExMZiZmXH27FkiIiK4f/++co3169djY2OjPFEIRYVcV6xYwbFjx7Czs+PEiRPUrl2bevXqkZKSwldffUVOTg4aGhro6+tz//59oqOjCQwMxMnJCTc3N3r06IGWlhZhYWGoqalx7do1dHV1GT9+PCdOnODw4cN06dJFuWZqaipHjhxh+/btSJL0WNskQRAEQXjVyiLYCgUsJUmqRVFQ5Qq4PbLPLmAA8F+gF/CnyNcqX506daJTp06PbS+tjEPxvh4eHgwYMICGDRuSkJDA559/jp2dHYGBgWhoaFCzZk3lmCtXruDu7l4icX3cuHH85z//4ciRI2RkZNC6dWvS0tK4fPkyoaGh/Pzzz9jb2zNx4kROnDiBsbExH374IUlJSQwYMIB79+5x5swZ2rVrR+XKlRk1ahTJyclK+x9ra2tu375dYux6enpoaWkxffp0WrZs+dT+i4IgCILwKrx0BXlZlvMBTyAIuARslWX5oiRJPpIkFTedWwNUkiTpKjAGmPCy1xVev5UrV9KiRQsAbGxsiIyMxNXVlblz5+Li4oKuri5paWksX76c5ORkJEli5cqVJCUl4enpSX5+PhMnTqR169bs37+fSZMmYW1tTcOGDalUqZLS1HvEiBF8+GHRYqm9vT1OTk60adOGtWvXIssyN27cYODAgaxevZomTZrw448/4uTkVOKWZTE1NTXWr19P+/btOX78uOi3JwiCILx2ZZKzJcvyXmDvI9umPPQ6B+hdFtcS3gwmJiaoq6srvSRtbGxISEjg/Pnz6OnpUatWLRYuXMiCBQswMjLiwIED/PzzzwwePLhEfpdKpXpqU297e3vOnj3L7du3adOmDZMnT0ZDQ+O5m0dnZWWRk5PDJ598gq2trWg6LQiCILx2b1yCvPB2GDp0KNWrVycyMpK5c+eyYsUKDAwMWLduHa1bt8bExISQkBBatGhBo0aNKCgoKFHN/WEPN/Vu1KgRe/fuJSkpicOHD2NnZ8dPP/2Evb09KpUKbW1tYmJisLOze2qQViwrK4sxY8aQl5eHLMuMGTOmrL8KQRAEQXgqEWwJ/1rdunU5d+4cFhYW2NnZERMTw5EjRzA0NKRx48YkJiaybt26EsfEx8eXeq5p06Yxa9YscnJyMDMzU24pmpqaIsuy0uPQ3d2dAwcOYGBggIGBQYknC/v166e89vb2Vl5v2LChrKYsCIIgCC9MBFsCUBQEjRw5Emtray5fvkzt2rXx8fEhMjKSxYsXU1BQQP369ZkwYYJSusHKygpfX18iIyNJS0tj/fr1pKWlsWnTJipVqkR8fDyffvopQUFB5OXlMX36dKKjo1GpVGzZsgUXFxdycnI4fvw4hw4dwtbWlkmTJiFJUolg6eH2OwMHDmTgwIGv++sRBEEQhH/tpRPkhXfHjRs36N27N9u2bUNPT49Nmzbh7e3NnDlz2LJlC/n5+UrFdygqOpuSkoK+vj4Anp6eVKlShebNmzNx4kR++OEHYmJi6NOnD46Ojly4cIHNmzcTEBBA586dgaJG1Bs2bGDr1q3k5uY+1qhaEARBEN52ItgSFFWrVsXW1haAzp07c/r0aczMzJSCot26dePs2bPK/iqVipCQkBIlH+rXr8+OHTsA+Oijj3BwcGDz5s20bNmSCRMmKE2mDQwMAAgLC2PAgAG4uLgQGhrKtWvXXsdUBUEQBOG1EcHWO664rlR8fDz79+9/6r6SJJV4X7xi9ark5eUxZ84c5s2bx5YtW/jiiy+e2VBaEARBEN42Ith6T9y+ffuZwVZxGQeA/fv3U69ePeLj47l5s6j15d69e5XE9dJUqFBB6XX4qI8//pjt27crdbDS0tLIzc0FwNDQkKysLA4dOvTC8xIEQRCEN51IkH9PLFu2jNjYWNzc3OjWrRtubo8W+YcaNWoQGBiIj48PtWvXxs3NjQYNGjB+/HglQf7LL7984jUqVqyIra0tzs7OfPLJJ/Tu/f+l1T7//HNu3LiBq6sr6urqfPHFFzg7O/P555/j7OxMpUqVlBZBgiAIgvAukd7krjkODg5yWFhYeQ/jrdaqVSuOHTtGeHg4fn5+pbbjgaLbjF5eXqJJsyAIgiD8S5Ikhcuy7PDodnEbURAEQRAE4RUSwZYAFBUPFatagiAIglD2RM7We0JXV/eJyevlpWnTplhYWCDLMiqVivHjx2NjY/PUY/z9/enZsyfa2tqvaZSCIAiC8HLEytZ7wtLSEpVKRZ8+ffD39y/v4QCgpaWFv78/mzdvxtPTk+XLlz/zmM2bN5OTk/MaRicIgiAIZUOsbL3jiiuyq6ur4+vrW86jebLMzEylrtejyfzz5s3D2tqazMxMEhMT8fDwwNDQkJUrV5bnkAVBEAThuYhgSyg3ubm5uLm5kZubS1JS0jODQVdXVzZt2sTKlSsxNDR8TaMsX7t37+bSpUt8//33rFq1Ch0dnRINt5+l+GlUQRAEofyIYEsoN8W3EQEiIyOZOnUqW7Zsea5ji/O98vPzUVNTo2vXrri5uaFSle2d8aFDh+Ll5UW9evVKbH84CBIEQRCEpxHB1iu2YcMGNDU1cXV1ZeHChVy5cgVfX19CQ0P5/fffqVChAlFRUeTm5tK+fXs8PDwIDQ0lICCABQsWAHDq1CkCAwOZP39+Oc/m1bGxsSElJYXk5GTU1NQoLCxUPiuuNP+whwO1+/fv88MPP5CZmYmHh8drG/PL2LNnD35+fkiShKWlJR06dGDNmjU8ePAAQ0NDZsyYgbGx8ROPv3XrFnPnziU5ORltbW1++OEHatasSXx8PJMnTyYrK4s2bdq8xhkJgiAITyKCrVesUaNGbNy4EVdXV6KiosjLyyM/P5+zZ89ib29Phw4dMDAwoLCwkGHDhhETE4ODgwNz5swhOTkZIyMjdu3aRffu3ct7Kq/U9evXKSgowNDQkAcPHhAbG0teXh65ubmEhoZiZ2cHlHyqMjs7mwkTJpCQkEBGRgYnT55k6NChPHjwgNmzZxMVFYWamhpjxozBwcGB3bt3ExISQk5ODrdu3aJt27aMGjUKQNn/4aD3Ubt27eLXX39FT0+POnXqoKmp+a/meu3aNdasWcPatWsxNDQkLS0NgF9//RVJkvjtt99Yv349o0ePfuI5Zs6cycSJEzE3N+fChQvMmTMHX19f5s+fT69evejataso5SG8kFmzZtG1a1elGb0gCGVHBFuvmLW1NZcuXSIzMxNNTU3q1q1LVFQUERERjB07loMHD7Jjxw4KCgpISkoiNjYWS0tLunbtyr59+3BycuL8+fP4+PiU91TKXHHOFoAsy0ybNg2VSkXVqlXp2LEjLi4umJqaYmVlpRzTs2dPPD09uXr1Kv/9738xMTFhyZIlQFF+UnJyMnv37kWSJLZs2cL169cZMWIEO3fuBCA6Ohp/f380NTXp2bMnLi4uVK1alREjRjwW9FpaWirXTUpKYuXKlWzcuBE9PT08PDyoW7fuv5p3aGgoHTp0UPLODAwMuHr1KhMnTiQpKYkHDx5gZmb2xOOzsrI4d+4cEyZMULYVN/COiIhg3rx5AHTt2pVly5b9qzEK758LFy6U+JkSBKHsiGDrFVNXV8fMzIzdu3djY2ODpaUlYWFh3Lx5E21tbfz8/NiwYQMGBgZ4e3srt8ycnJwYPXo0mpqadOjQATU1tXKeSdk7ffr0Ez8bNWqUsur0MBcXF1xcXGjVqhUWFhYsWrSIpUuX0qpVK+U7ioiIwMXFBYCaNWtSvXp1bty4ARTleunp6QFQu3Ztbt++TdWqVZ8Y9Ba7cOECjRs3xsjICIBOnToRFxdXNl8ERU9curu707p1a8LDw5/6pKUsy+jr6z+xhIckSWU2LuH9EBsbi7m5eZnnPAqCUET8l/UaNGrUCD8/P+zt7WnUqBHbt2/HysqKjIwMdHR00NPT4/79+5w8eVI5xsTEBBMTE9asWYOTk9NrGefAgQNf+Jjg4GCuXbv2CkbzbObm5mzatAkLCwvmz5/PnTt3lGCoNMuXLyc5OVl5r1KpKCgoID4+Hj8/P1asWEFAQAAtW7YsNU+srDRp0oRDhw6RmpoKQFpaGhkZGZiYmADwxx9/PPX4ChUqYGZmxqFDh4Ci4OvKlSsA2NnZERQUBMC+ffte1RSEd8zJkydp0aJFeQ9DEN5ZIth6Dezs7EhKSqJhw4YYGxujpaVFo0aNqFOnDlZWVvTq1YvJkyc/livRuXNnqlatSq1atV7LONeuXfvCxwQHBxMbG/sKRvNsiYmJaGtr07x5c7Kysvjwww+RJIlGjRopgUZcXBx37tyhRo0aTzzP04LeYg0aNODMmTOkpqaSn5+vBDr/Ru3atRk4cCBDhgyhT58+LFy4EA8PD8aPH4+7u/tzlbWYMWMGv//+O3369MHZ2ZmjR48CMHbsWAIDA3FxcSEhIeFfj1F4v/z3v/+lefPm5T0MQXhnSbIsl/cYnsjBwUEOCwsr72GUm3nz5mFlZUWPHj1ey/VatWrF4sWLSy0o6uTkxLJlywgJCUFNTY1mzZrRrl07vLy80NPTQ09Pj3nz5vHBBx+U6Zji4+MZOXIkDRs2JDIyknr16tG9e3e6du2KJEnKrcPCwkIaN25MtWrVmDhxIv7+/pw/f55Lly5RrVo1GjVqxJkzZ2jUqBELFizgr7/+YuDAgdSoUQMbGxvU1dW5dOkSVatWRU9Pj9atW+Pk5FSi9MPDCfJWVlZoaGi89tIPvr6+2Nvb07Rp0xLbHy0EKwjPKycnBw8PD9avX1/eQxGEt54kSeGyLDs8ul3kbL2h3N3d0dHRwcvLq7yHAkBqaipHjhxh+/btSJJEeno6+vr6tG7dmlatWtG+fftXdu2bN28yd+5cateuTf/+/dm/fz8JCQmEhISwa9cufHx80NbWRk1NjdOnT/PLL78wb948Nm3ahI2NDVOmTCEmJoa+ffsyaNAgUlJSWLNmDeHh4ejo6LB+/Xry8vKYOXPmY9detWqV8rp79+7l/lTosGHDyvX6wrsnLCwMB4fHfjcIglCGRLD1htq4cWOZnu9lK4nr6emhpaXF9OnTadmyJa1atSrD0T2dmZkZFhYWQNEtuCZNmiBJEhYWFty+fZuMjAymTp1KXFwckiSRn58PwJkzZ3B1dQWKekMWJ7yfP3+ea9euMWjQIAAePHhAw4YNX9t8HvVwCYuCggIGDx7MjRs3OHbsGDk5Odja2jJp0iQkScLb21sJbk+ePMmCBQvQ1tZWSmO8aUaNGsV//vMfJR9NePOcPHnylf5lSRAEEWwJj3hSQVE1NTXWr19PaGgohw8fZuvWra+t16KGhobyWqVSKfWtigOrFStW4ODgwPz584mPj39mYVNZlvn444+ZNWvWKx3383q0hEVGRgbNmjVjyJAhAEyZMoVjx47RunVr5Zi8vDxmzJiBr68vH374IRMnTiyXsT/L0qVLy3sIwjOcO3eOMWPGlPcwBOGdJhLkhRKqV6+uFBRNT08nNDQUKKrtlJGRwSeffMKYMWOUp990dXXJzMwszyGTkZFBlSpVgJJP8tnb27N//34A/v77b2JiYgBo2LAh586d4+bNm0DRylJZlnF4URYWFpw6dYqlS5dy9uxZ9PT0CAsLY8CAAbi4uBAaGvrYE5/Xr1/HzMwMc3NzJEmiS5cu5TR64W23adMm1NXF37sF4VUS/4W9R9LT09m/fz+9e/cu9XNJkigoKODWrVuPFRTNyspizJgx5OXlIcuy8jfhTz/9lBkzZhAQEPBKEuSfR//+/fH29mb16tW0bNlS2d6rVy+mTZtGr169qFWrFtbW1gAYGRnh7e3N5MmTlWKgw4cPx9zc/LWPHf6/hMXx48dZsWIFTZo0ITAwED8/P6pWrcqqVauUcQqCIAhvHxFsvUfS09MJDAwsNdhKTU3FwMAAKMpvKq3Vy4YNGx7bZmtrS2BgYNkP9n9MTU1LjMXb27vUz3bs2KFsHz58OFDUP/FJtwqbNGlS6nzKQ2JiIhUrVqRLly7o6+vz22+/AWBoaEhWVhaHDh2iQ4cOJY4p7oN469YtPvjgA2UF700jcrYEQRBEsPXOGzVqlPKU3fLly7l16xZubm44ODgQExNDWloaJ06coFKlSo+VMTh27Bg+Pj4sWbIEAwODUhsfCy/v6tWrLFmyBJVKhbq6OhMmTCA4OBhnZ2cqVapE/fr1HztGU1OTyZMn8+2336KtrU2jRo2UnpFvEpGzJQiCQFGy8Jv6p3HjxrLw7xUWFsoFBQWyLMtyy5Yt5X/++Ufu3bu3LMuynJ+fL2dkZMiyLMv9+/eX27VrJxcWFir7XL9+XW7durU8duxYWZZlediwYfKNGzdkWZbl8+fPyx4eHi80liNHjsiNGzeWY2NjX+i4sLAw+dtvv32hYwRBEAShPABhcinxjFjZesfEx8fj6elJgwYNuHz5MteuXVOqnfv7+3Py5EkGDRqEiYkJ169fB4rq7BQUFODq6kpaWhopKSl4eXmhUqk4e/Yszs7OXLhwQWkvA7xwDlFQUJDSSuZZTwsKgiAIwrvkpYItSZJ+BJyAPOBv4GtZllNK2e86kA4UAPlyKdVVhbJz8+ZNpk2bRsOGDZW+iitXrmTSpEk0a9aMZcuW0a5dO0xMTNi5cyfDhw/nxIkTLFq0iNDQUEaPHo29vT1169bFyMiIESNG0LNnzyc2Pn6WrKwsIiIiWLlyJV5eXnh4eCjNlg0NDfn777+xtrZm+vTpSJL0VtSPEgRBEITn9bKlHw4CDWRZtgGuAE8r9tNWlmU7EWi9etWrV3+sSGdERASOjo7k5OSgq6uLhYUFFSpUQF1dncTERGU/CwsLcnNzmT9/PmFhYURHRz+18fHzOHr0KM2bN8fc3BxDQ0MuXboEQHR0tNLL759//uHcuXNK/ahFixaxceNG7t27VwbfiCAIgiCUn5cKtmRZPiDLcv7/3v4FvP7n/oXHaGtrl7pdR0cHW1tbnJ2dSUxMJD4+HhcXF+Li4jAzMwOKiobKsoyOjg6DBw8mIiKCkJCQJzY+fh5BQUF8+umnAHTq1ImgoCAA6tevT5UqVVCpVNSpU4f4+HhRP0oQBEF455RlztZAYMsTPpOBA5IkycBKWZZXPWE/JEkaCgwFyq3u0bvIzs6OmTNnsm7dOvLz83F3d8fd3Z1+/fopzZZNTU3R1dWlcePGAJiYmNCtWzelcvmyZcte+LppaWmEhoZy9epVpY6XJEm0bNlSqQQPRRXqCwoKymaygiAIgvAGeebKliRJhyRJulDKnx4P7TMZyAc2PeE0LWVZtgc6AyMkSWr9hP2QZXmVLMsOsiw7iNo8ZadevXq0adMGV1dXRo0ahYWFBXp6ek89xsHBgdjYWNzc3Dhw4MC/uu6hQ4fo2rUrf/zxB7t372bv3r2YmZlx9uzZUvd/uH4U8MbWjxIEQRCE5/XMlS1Zljs87XNJkr4CugHt//fYY2nn+Od//0yQJGkn0BQIeeHRCs/0aBHQ3bt3K6/d3d0ZOnQoOTk5DBkyRKmovmrV/y80GhoaKscYGBi8dOHPoKAgBgwYUGJbu3bt2LZtW6nV5t+W+lGCIAiC8LykJ8RHz3ewJH0GLATayLKc+IR9KgAqWZbT//f6IOAjy/IzlywcHBzksLCwfz0+oaTJkycTGxtLbm4u3bp14+uvvy7vIQmCIAjCO0OSpPDSHgR82WDrKqAFFD8y9pcsy8MkSTIFVsuy3EWSpNrAzv99rg74y7I883nOL4ItQRAEQRDeFk8Ktl4qQV6WZYsnbI8Huvzv9TXA9mWuIwiCIAiC8LZ62TpbgiAIgiAIwlOIYOs9EBwczLVr18p7GIIgCILwXhLB1nsgODiY2NjY8h6GIAiCILyXRCPqt9Tq1avZu3cvRkZGVK1aFWtra9q2bcvcuXNJTk5GW1ubH374gbS0NEJCQjhz5gxr1qxh3rx5pZZcEARBEATh1RDB1lsoKiqKP//8k4CAAPLz8+nbty/W1tbMnDmTiRMnYm5uzoULF5gzZw6+vr60bt2aVq1a0b59+/IeuiAIgiC8d0Sw9RaKiIigTZs2aGpqoqmpSatWrcjLy+PcuXNMmDBB2S8vL68cRykIgiAIAoicrXdGYWEh+vr6+Pv7K3+2bdtW3sMS3jPp6ekEBgY+dZ/4+HicnZ1L/Wzo0KFERUW9iqEJgiCUGxFsvYXs7OwICQkhLy+PrKwsjh8/jra2NmZmZhw6dAgAWZa5cuUKALq6umRmZpbnkIX3xPMEW4IgCO+b/2vv3qO6qvL/jz83KEoi3jJT/Jb6QwIUEcRLF/PraGYta8oUEbLEfuLM0jSbyhxLMNG+w5glWbo0WypJpDbldan5LZc6MxoEWoZpDnkBLHUlqWGisn9/QJ+fF+iinw+Hy+uxFotz2Z+z3+dw1se3Z++zt5oRa6BLJ5Vu3ry5a1Lp5ORkXn75ZRYtWsSFCxcYMGAAQUFB3HvvvSQnJ5ORkaEO8uJRc+fOJT8/n9jYWHr27AnAP//5T4wxPPHEEwwYMOCy8ufOnWPatGns37+fdu3ace7cOSfCFhHxKCVbNVRFk0q3adOG119//aqy4eHhetogVWLcuHEcOHCA9PR0Pv74Y1auXElGRgZFRUWMGDGCyMjIy8qvXLmShg0bsnLlSr7++mvi4uIcilxExHOUbNVQM2bMuGxS6eDgYKdDErnMrl27GDhwIF5eXjRv3pxu3bqRm5tLYOD/n+UrOzubmJgYADp27EjHjh2dCldExGOUbNVQM2b8prm8RURExGHqIF/NjB8/ntOnTwPQu3dv4Jff3hKpTho1akRxcTFQ9iLHpkDjxjcAABhOSURBVE2bKC0t5eTJk2RnZ9OpU6fLykdGRrJhwwYA/vOf//D1119XecwiIp6mJ1vVTGpqqtMhiFyzJk2aEB4eTnR0NHfeeScdO3YkJiYGYwzjx4+nRYsWFBYWusoPGTKEadOmMWTIENq3b09ISIiD0YuIeIaSrSq2dOlSfHx8iImJYfbs2ezfv5/58+eTmZnJqlWr2L17N2lpaTRt2tTpUEWuyZVN3BMmTLhsvU2bNixfvhyABg0aMHPmzCqLTUTECWpGrGIRERHk5OQAZdPuFBcXc+HCBXJycq56U0tERERqPiVbVSwkJIS9e/fy448/4uPjQ5cuXcjNzWXXrl107drV6fBERERqrXPnzpGQkEBpaSkA+fn5vPTSS0RHR/Poo48ye/ZsTp065fZ6lWxVsXr16hEQEMCaNWvo0qULERERZGVlceTIEdq3b+90eCIiIrXWqlWr6Nu3L15eXuzZs4fnn3+ee+65h4yMDJYuXUp4eDhPPvkkP/zwg1vrVbLlgIiICNLS0oiMjCQiIoL333+f2267DWOM06GJiIjUWhs2bKBPnz6UlpaSkpLCq6++yu23346XlxdeXl7069ePsWPHMn/+fLfWqw7yDujatSuLFi0iLCwMX19fGjRoQEREhNNhiYiI1Frnz58nPz+fNm3asGPHDnr16kXLli358MMPWb58OcHBwZw/f57p06ezcOFCt9atZMsBPXr0YOfOna71f/zjH67lNWvWuJa3bdsGXP72loiIiPx+RUVFNG7cGID9+/fTuXNnTp48yfr161m8eDEHDhxg8uTJANx4442cPHmSZs2auaVuNSO62enTpx2dh3D+/Pl8+umnjtUvIiJSHTVo0ICSkhLXure3NwUFBYSFheHj40NoaKhr2KVTp07h7+/vtrqVbLmZ08nWn/70J3r06OFY/SIiItWRv78/Fy9epKSkhMDAQPbs2UNAQABffPEFJSUlfPXVVxQVFZGZmUnLli3x9vZ2W91qRnSzuXPnkp+fT2xsLD179rxqQMdr8dZbb7F+/XqaNWtGq1atCAkJoWfPnsycOZOffvqJtm3bMnXqVPz9/UlKSqJ3797069ePBx54gEGDBrF161YuXLjA3/72N9q1a8fJkyeZMmUKJ06cICwsjJ07d/LOO+9oIFUREanVevXqxa5du+jRowfz5s1j6NChDBw4kJEjRxIcHExgYCAff/wxzz33nFvr1ZMtNxs3bhxt27YlPT3dLYlWbm4uH3/8MRkZGbz++uvs3bsXgKlTpzJ+/HgyMjIIDAystDNf06ZNWbZsGUOGDCEtLQ2AhQsX0r17d5YvX06/fv349ttvrztOpx08eJBRo0YxbNgwEhISKCoqcjokERGpZqKjo1m7di316tVj4sSJTJw4kaCgINLT03nhhReIi4vjnnvu4YYbbnBrvUq2qrldu3bRp08ffHx8uOGGG+jduzdnz57l9OnTrhHnBw0aRHZ2doWf79u3L1A2mOrRo0ddx7z33nsBuOOOO9zaLn09zp49y4QJExg+fDjR0dFs2rSJhQsX8thjjxEdHc2MGTOw1pKfn09cXJzrc4cPH+app55i+vTpvPfee3Tp0oX333/fwTMREZHqKDg4mKioKEpLS4mMjCQpKYmMjAxiY2OJi4vjk08+oUOHDm6vV82ItZyPjw8AXl5eXLhwweFoftm///1vWrZsyZw5cwA4c+YMvXr1YvTo0UDZ07xt27Zx99134+fnx/79+wkKCmLNmjUMHz6cgIAAAEpKSmjSpIlj5yEiItXXgw8+6Fpu3749ycnJHq9TT7bcrFGjRhQXF7vteF27dmXr1q2UlJRQXFzM9u3b8fX1xd/f3zXH4vr16+nWrdtvPmZ4eDgfffQRADt27PDI1ATXIjAwkJ07d5KamkpOTg5+fn5kZWXx+OOPM2zYMDIzM8nLywPgoYceYvXq1ZSWlrJp0yYGDhwIlCVs//rXv3jooYecPBUREREXPdlysyZNmhAeHk50dDR33nnndffbCg0NpU+fPsTExNC8eXMCAwPx8/Nj2rRprg7yAQEBJCYmVnqMNWvWsGXLFtf66NGjmTJlCuvWraNLly60aNHC7e3T1+KWW25h2bJlbN++nXnz5tG9e3dWrFhBWloarVq1YsGCBa7Xdvv16+fqexYSEkKTJk0oLS1l+vTpzJ8/3zWWioiIiNOUbHnAjBkz3Hq8Rx99lISEBH766SdGjx5NSEgIQUFBLF68+KqySUlJruVLB0ht1aoVr7zyCgB+fn7MnTsXb29vPv/8c3Jzc13NjU46fvw4TZo04f7776dx48Z8+OGHQFkn/+LiYjZv3kz//v2BsubRXr168fLLLzN16lQATpw4gZ+fH7fccotj5yAiInIlJVs1wIwZM/jyyy/JzMwkKiqKv/71r4SGhvLggw8yf/58Tp486WpznjVrFufOnaNhw4YkJiZy6623Xnas7du3k5qaSmlpKT/88APffPMNwcHBTJo0icTEREefcB04cIA5c+bg5eVFvXr1eP7559myZQvR0dG0aNGCTp06XVb+vvvuY8uWLfTq1QuAxo0b89RTTzkRuoiISKWMtdbpGCoVFRVls7KynA6jWigsLOShhx4iPT2dDh068NhjjxEUFMSLL77I1q1bWb16NS+99BINGzbE29ubTz/9lJUrV5KSksKaNWvYu3cv3bt3Z9myZcyePZvS0lKeffZZUlNT8fX1ZcmSJZSUlLg6o9cEaWlpnDlzhj//+c9A2ZOxv//976SkpDgcmYiI1EXGmM+stVFXbr+uJ1vGmCRgNHC8fNNfrbXrKyg3EJgDeANvWWv/53rqrasCAgIIDAwEoEOHDnTv3h1jDIGBgRw9epQzZ86QmJjI4cOHMcZc9vZhZmYmubm5vPHGGzRq1Iht27aRl5fHE088AZRN0BkWFubIeV2LZ555hvz8/MtmZm/ZsqUSLRERqXbc0Yz4qrV2VmU7jTHewBvAPUA+kGmMWW2tzXVD3XVK/fr1XcteXl6uflY/J1bz5s0jKiqKWbNmUVhYyJgxY1zl27ZtS0FBAYcOHSI0NBRrrWsU+ppo1qxKbzkREZFqpSqGfugBHLDW5llrS4AM4I9VUG+dc+bMGW666SYA1q5de9m+1q1bk5KSQmJiInl5eYSFhbF7926OHDkClA0oevjw4SqPWUREpLZzR7I1zhjzuTHmbWNMswr2BwBHLlnPL98mbvbYY48xd+5cYmNjKxzAtF27diQnJzNp0iR+/PFHkpKSmDJlCjExMcTHx3Pw4MGqD1pERKSW+9UO8saYzcDNFeyaAuwATgAWmA60ttaOuuLzQ4CB1tr/W74+AuhprR1XSX0JQALALbfc0u3QoUO/64REREREnHDNHeSttf1/YwULgbUV7CoA/uuS9bbl2yqrbwGwAMreRvwtdYuIiIhUV9fVjGiMaX3J6sPAngqKZQIdjTHtjTE+QAyw+nrqFREREakprvdtxBRjTFfKmhEPAmMAjDFtKBvi4X5r7QVjzDhgI2VDP7xtrf3yOusVERERqRGuK9my1o6oZHshcP8l6+uBq8bfEhEREantqmLoBxEREZE6S8mWiIiIiAcp2RIRERHxICVbIiIiIh6kZEtERETEg5RsiYiIiHiQki0RERERD1KyJSIiIuJBSrZEREREPEjJVi1graW0tNTpMERERKQC1zs3ojiksLCQcePG0blzZ7766iv69+/P9u3bKSkpoW/fvowZM8bpEEVERAQ92arRjhw5wtChQ3n66ac5fvw4S5YsIT09nb1795Kdne10eCIiIoKSrRqtdevWhIWFsWPHDnbs2EFcXByPPvooBw8e5MiRI06HV6nevXsDcPz4cZ577jkA1qxZQ0pKipNhiYiIeISaEWuwhg0bAmV9tuLj4xk8eLDDEf0+LVu2VIIlIiK1np5s1QK33347q1atori4GIBjx47x/fffOxzVryssLCQ6Ovqq7du3byc+Pp6ioiJ27NhBfHw8cXFxTJo0yXWOIiIiNYWebNUCvXr14ptvviE+Ph6AG264genTp9O8eXOHI/v9PvnkE5YtW8acOXMoLS1l0aJFvPnmm/j6+rJkyRKWLVvG6NGjnQ7ToxYsWICvry8jRoxwOhQREXEDJVs1VJs2bVi+fLlrffjw4QwfPvy6jnn69Gk2bNjA0KFD+eyzz0hLS+O11167qtz06dOJi4ujQ4cOlR4rKSmJ3r17069fv99cf2ZmJrm5ubzxxhs0atSIbdu2kZeXxxNPPAHA+fPnCQsL+/0nJiIi4iAlW+Jy+vRpVqxYwdChQ3+x3IsvvuiR+tu2bUtBQQGHDh0iNDQUay09e/Zk5syZHqlPRESkKqjPlrjMnTuX/Px8YmNjmTNnDsXFxTz33HM88sgjvPDCC1hrAUhISCA3Nxcoe7PwzTffZPjw4YwcObLCvmLz5s0jKSnpVwdebd26NSkpKSQmJpKXl0dYWBi7d+92vVl59uxZDh8+7OazFhER8SwlW+Iybtw42rZtS3p6OhMmTGDfvn0888wzrFixgoKCAnbv3n3VZ86ePUvnzp159913iYyM5IMPPrhs/5w5czh58iSJiYl4ef367dauXTuSk5OZNGkSP/74I0lJSUyZMoWYmBji4+M5ePCgu05XRESkSqgZUSrVqVMnbrrpJgCCgoIoLCyka9eul5WpX7++a9ys4OBgdu7c6dr31ltv0blzZ6ZMmXLZZ7Zt2wZc3u/sgQce4IEHHgDgtttuY8WKFUBZ0+LSpUs9cHbVV0JCgtMhiIiIG+nJllTKx8fHtezt7c3FixevKlOvXj2MMRWWCQ0NZe/evZw6dcrzwYqIiFRTSrbEpVGjRm4dx+qOO+5g5MiRTJgwQeNj/Q7vv/8+69atczoMERFxEzUjikuTJk0IDw8nOjqahg0bumWcrv79+1NcXMzEiRNJTU2lQYMGboi0dnvkkUecDkFERNzI/PyGWXUUFRVls7KynA5DKlBYWMhTTz112VhfIiIidZkx5jNrbdSV29WMKCIiIuJBSrbkuhUUFBAbG8vSpUt59tlnefLJJ3n44YdJTU11ldm4cSPDhg0jOjratX3z5s3Mnj0bgHfffZcHH3zQdbxRo0ZV/YmIiIh4gPpsyXU5dOgQkydPJikpiX379rFv3z7S09Px8fFh8ODBDBs2DC8vL1JTU3nnnXfw9/dn7NixbNmyhYiICNewDjk5OTRt2pRjx46Rk5NDZGSkw2cmIiLiHnqyJdesqKiIp59+muTkZIKCggDo0aMHfn5++Pj40KFDB44ePUpubi7dunWjWbNmeHt7c99995GdnU2LFi0oLi6muLiY7777joEDB5KTk0NOTg4REREOn52IiIh7KNmSa+bn58fNN9/Mrl27XNsuHZvLy8urwrG5LtWlSxdWr17NrbfeSteuXcnJyeGLL74gPDzcY3GLiIhUJSVbcs3q1avHrFmzWLduHRs2bKi0XKdOncjOzqaoqIjS0lI2btxIt27dAIiIiCAtLY3IyEiCg4PJysqifv36+Pn5VdVpiIiIeJT6bMl18fX15bXXXmPs2LHcf//9FZa58cYbefLJJxkzZgzWWu666y769OkDlCVb3333HZGRkXh5edGqVSvatWtXhWcgIiLiWRpnS0RERMQNKhtn67qebBlj3gNuK19tChRZa7tWUO4gcBq4CFyoKBARERGR2ui6ki1r7bCfl40xrwA//ELxvtbaE9dTn4iIiEhN45Y+W8YYA0QDf3DH8URERERqC3e9jdgb+M5a+3Ul+y2wyRjzmTEm4ZcOZIxJMMZkGWOyjh8/7qbwRERERJzxq0+2jDGbgZsr2DXFWruqfHk48O4vHOYua22BMeYm4CNjzFfW2q0VFbTWLgAWQFkH+V+LT0RERKQ6+9Vky1rb/5f2G2PqAYOBbr9wjILy38eMMR8APYAKky0RERGR2sQdzYj9ga+stfkV7TTGNDLGNP55GRgA7HFDvSIiIiLVnjuSrRiuaEI0xrQxxqwvX20FbDfG7AY+BdZZaysfblxERESkFrnutxGttSMr2FYI3F++nAdoojsRERGpkzQ3ooiIiIgHaW7EGmjdunWkpaVhjKFjx47079+fRYsWcf78eZo2bUpycjLNmzdnwYIFfPvtt+Tn5/Ptt98SGxtLTExMlcW5YMECfH19GTFiRJXVKSIiUt0o2aph8vLyWLRoEW+//TZNmzbl1KlTACxevBhjDB9++CFLlixh4sSJABw8eJD58+dTXFzM4MGDGTJkCPXq6c8uIiJSVfSvbg2TmZlJ//79adq0KQD+/v4cOHCAyZMnc+LECc6fP09AQICr/F133YWPjw8+Pj40b96c77//nptuuslj8b399tusXbuWZs2a0apVK0JCQti/fz8zZ87kp59+om3btkydOhV/f38SEhLo3LkzWVlZnDlzhhdffJGIiAiPxSYiIuIE9dmqBVJSUhg2bBjvvfceU6ZM4dy5c6599evXdy17eXlx4cIFj8Wxd+9eNm7cSHp6OqmpqeTm5gIwdepUxo8fT0ZGBoGBgSxcuND1mYsXL7J06VL+8pe/XLZdRESktlCyVcN0796dzZs388MPZXN+nzp1ijNnztCyZUsA1q5d61hsOTk59O3bl4YNG9KoUSPuvvtuzp49y+nTp4mMjARg0KBBZGdnuz7zhz+UTacZEhJCYWGhI3GLiIh4kpoRa5gOHTowatQoRo8ejbe3N7fddhtjxoxh0qRJ+Pv70717dwoKCpwO8zf7+cmbl5cXFy9edDgaERER91OyVQMNGjSIQYMGXbatT58+V5VLSLh8zu/ly5d7NK7IyEiSkpKIj4/n4sWLbNu2jcGDB+Pv709OTg4RERGsX7+ebt0qndlJRESk1lGyJW4THBzMgAEDGD58OM2aNSM0NBSAadOmuTrIBwQEkJiY6HCkIiIiVcdYa52OoVJRUVE2KyvL6TBEREREfpUx5jNrbdSV29VBXkRERMSDlGyJiIiIeJCSLREREREPUrIlIiIi4kFKtkREREQ8SMmWiIiIiAcp2RIRERHxICVbIiIiIh6kZEtERETEg5RsiYiIiHiQki0RERERD1KyJSIiIuJBSrZEREREPMhYa52OoVLGmNPAPqfjqMNuBE44HUQdpuvvLF1/Z+n6O0vX/9rcaq1teeXGek5E8jvss9ZGOR1EXWWMydL1d46uv7N0/Z2l6+8sXX/3UjOiiIiIiAcp2RIRERHxoOqebC1wOoA6TtffWbr+ztL1d5auv7N0/d2oWneQFxEREanpqvuTLREREZEarVomW8aYocaYL40xpcaYqCv2TTbGHDDG7DPG3OtUjHWFMSbJGFNgjNlV/nO/0zHVdsaYgeX39wFjzPNOx1PXGGMOGmO+KL/fs5yOp7YzxrxtjDlmjNlzybbmxpiPjDFfl/9u5mSMtVkl11/f+25WLZMtYA8wGNh66UZjTCgQA3QCBgJvGmO8qz68OudVa23X8p/1TgdTm5Xfz28A9wGhwPDy+16qVt/y+12vvnveYsq+zy/1PPC/1tqOwP+Wr4tnLObq6w/63neraplsWWv3WmsrGsz0j0CGtfactfYb4ADQo2qjE/GoHsABa22etbYEyKDsvheplay1W4Hvr9j8R2BJ+fIS4KEqDaoOqeT6i5tVy2TrFwQARy5Zzy/fJp41zhjzefnjZj3O9yzd486zwCZjzGfGmASng6mjWllrj5Yvfwu0cjKYOkrf+27kWLJljNlsjNlTwY/+F1/FfuVvMQ/4P0BX4CjwiqPBinjeXdbaSMqacscaY+52OqC6zJa9Mq/X5quWvvfdzLHpeqy1/a/hYwXAf12y3rZ8m1yH3/q3MMYsBNZ6OJy6Tve4w6y1BeW/jxljPqCsaXfrL39K3Ow7Y0xra+1RY0xr4JjTAdUl1trvfl7W97571LRmxNVAjDGmgTGmPdAR+NThmGq18i+6nz1M2csL4jmZQEdjTHtjjA9lL4SsdjimOsMY08gY0/jnZWAAuuedsBp4vHz5cWCVg7HUOfred79qORG1MeZh4HWgJbDOGLPLWnuvtfZLY8xyIBe4AIy11l50MtY6IMUY05Wyx/gHgTHOhlO7WWsvGGPGARsBb+Bta+2XDodVl7QCPjDGQNn3Y7q1doOzIdVuxph3gf8GbjTG5AOJwP8Ay40xTwCHgGjnIqzdKrn+/63vfffSCPIiIiIiHlTTmhFFREREahQlWyIiIiIepGRLRERExIOUbImIiIh4kJItEREREQ9SsiUiIiLiQUq2RERERDxIyZaIiIiIB/0/20Jy9RojXkYAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rC_kUNksdI37"
      },
      "source": [
        "In fake news articles, we see a much more skewed distribution. \n",
        "\n",
        "The politicia names no longer appear as clearly clustered (`Donald` outside of main \"blob\"), and some words appear more \"extreme\" in their use (`country`, `people`, `government`). \n",
        "\n",
        "The \"subjective\" words that we noted in the topic analysis (`think`, `know`) seem to form their own cluster as well."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QHhffW3VdI38"
      },
      "source": [
        "### Exploration\n",
        "Now it's time for you to try out some more tests with word embeddings!\n",
        "\n",
        "- Increasing the **window size** when training embeddings can help the embeddings capture more global context (e.g. associating `tomato` with cooking details from the wider sentence context). How would this help capture divides between fake news and real news?\n",
        "- One way to determine the **connotation** of a word in embedding space is to look at its proximity to positive and negative words: e.g. if `Barack` is consistently closer to words like `bad` and `terrible` than to `good` and `nice`. Can you come up with a way to test word connotations using this kind of approach, and determine whether some words have consistently better or worse connotations in fake news articles?\n",
        "- Another useful aspect of word embeddings is their tendency to **cluster** words into general semantic fields, e.g. grouping all politician names near one another. Using the visualization technique from above, try to find words that (1) consistently fall into neat clusters and (2) sometimes appear outside of the expected clusters in the data. Which political and organizational words tend to be represented outside of their expected cluster, and why do you think that happens? "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Vkx2KQNXncQL"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}