quokka

1. Download data

   Download files from these links and copy them to the data directory

   Energy Hub

   Energy Hub Training set - 
   
   Energy Hub Validation set - 
   
   Energy Hub Test set - 
   

   Reuters

   Reuters Training set - 
   
   Reuters Validation set - 
   
   Retuers Test set - 
   

2. Downloading Necessary Packages
   • Download NLTK stopwords using

     import nltk
     
     nltk.download('stopwords')
     

   • Download Mallet from here. Unzip and copy it to the directory.

     If you use Google Colab:

     !wget http://mallet.cs.umass.edu/dist/mallet-2.0.8.zip
     !unzip mallet-2.0.8.zip
     

   • Download GloVe embeddings from here. Unzip and copy it to the directory.

     If you use Google Colab:

     ```
     !wget https://nlp.stanford.edu/data/wordvecs/glove.6B.zip
     !unzip glove*.zip
     ```
     

3. Build Topic-Entity Triples

   This step involves

   • Training a Topic Modeler over the corpus
   • Extracting Named-Entities using spaCy
   • Building Triples using Dependency parser and POS tagger
   • Apply Topic Entity Filter over these triples

   Run the following python file.

   python data_preprocess.py <dataset>

   Change <dataset> to "energy hub" or "reuters" to select the corpus.

4. Training Models

   Run the following python file.

   python train.py <dataset> <model>

   Change <dataset> to "energy hub" or "reuters" to select the corpus.

   Change <model> to the following options

   • text - for GloVe based text model
   • topics - To use topic distributions
   • entites - To use Glove-enriched named entities
   • triples - To use Glove-enriched triples
   • text_topics - To use text and topic distributions
   • text_triples - To use text(GloVe) and triples(GloVe)