This is a course project done for Department of Computer Engineering, Delhi Technological University (formerly, Delhi College of Engineering) for a course in Machine Learning (Course Code - CO327) for the Fall Semester 2021.
- Fake news existed way before social media but was multifold when social media was introduced.
- Fake news is a piece of news designed to spread hoaxes, propaganda, and disinformation deliberately.
- Fake news stories are usually spread through social media sites like Facebook, Twitter, etc.
- People are profiting by publishing fake news online.
- In general, the goal is profiting through clickbaits.
- Such clickbaits lure users and entice curiosity with flashy headlines or designs to click links to increase advertisement revenues.
Some of the major problems related to fake news that are generally faced by people are:
- By clicking on a clickbait, users are led to a page that contains false information.
- Fake news influences people’s perceptions.
- The risk of fake news has become a global problem that even major tech companies like Facebook and Google are struggling to solve. It can be difficult to determine whether a text is factual without additional context and human judgement.
- This project aims to develop a method for detecting and classifying news stories using
Natural Language Processing. - The main goal is to identify fake news, which is a classic text classification issue.
- We gathered our data, pre-processed the text, and translated our articles into supervised model features.
- Our goal is to develop a model that classifies a given news article as either fake or true.
There are two delimitations associated with our project:
- Our system does not guarantee 100% accuracy.
- The system is unable to test data that is unrelated to the training database.
There are basically two types of fake news:
- Visual based type: Visual based are mainly photoshopped images, and videos which are posted on various social media platforms.
- Linguistic based type: Linguistic based are mainly the manipulation of text and string content. These issues are with blogs, news, and emails.
- There have been attempts made to use Artificial Intelligence techniques specifically machine learning/deep learning technique and natural language processing (NLP), to automate the process of detecting fake news and stopping it from spreading any further.
- It can be possible to teach a computer how to read and differentiate between real news and fake news using Natural Language Processing (NLP).
- The building blocks of our project are Dataset and Machine Learning.
- It has the following attributes:
- id: unique id for a news article
- title: the title of a news article
- author: author of the news article
- text: the text of the article; could be incomplete
- label: a label that marks the article as potentially unreliable
- The size of the dataset is 7796 x 4, which means it has 7796 rows and 4 columns.
- The dataset is published on Kaggle for research and educational purposes.
- We have 80% of data as training data and 20% data as testing data.
The steps involved in the workflow of our project are:
- First, we import the necessary libraries required for our project.
- Then, we try to read the data into a DataFrame and get the shape of the data and the first 5 records.
- After this, we get the labels from the DataFrame.
- Now, we split our dataset into training and testing sets. The training set would contain the data which will be fed into the model. The testing set will contain the data on which we test the trained and validated model.
- Now, we initialize a
TfidfVectorizerwith stop words from the English language and a maximum document frequency of 0.7 (terms with a higher document frequency will be discarded). - Next, we’ll initialize a
PassiveAggressiveClassifierand try to calculate the accuracy. - Finally, we print out a confusion matrix to gain insight into the number of false and true negatives and positives.
Fig: ML Model Architecture
- We used TF-IDF for feature extraction.
- We trained our data by using PA algorithm.
- We tested the efficiency of the classifier using accuracy metric.
- It denotes Term Frequency and Inverse Document Frequency.
- TF (Term Frequency): The number of times a word appears in a document is its Term Frequency. A higher value means a term appears more often than others, and so, the document is a good match when the term is part of the search terms.
- IDF (Inverse Document Frequency): Words that occur many times a document, but also occur many times in many others, may be irrelevant. IDF is a measure of how significant a term is in the entire corpus.
- The TF-IDF is very commonly used in data mining and data recovery.
- Search engines commonly use TF-IDF to rate and rank documents.
- TF-IDF may be used to separate stop-words in a variety of subjects such as text summarization and classification.
𝑇𝐹(𝑥) = (𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑖𝑚𝑒𝑠 𝑤𝑜𝑟𝑑 𝑥 𝑎𝑝𝑝𝑒𝑎𝑟𝑠 𝑖𝑛 𝑎 𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡)/(𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑤𝑜𝑟𝑑𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡)
𝐼𝐷𝐹(𝑥) = ln((𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡𝑠)/(𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡𝑠 𝑤𝑖𝑡ℎ 𝑤𝑜𝑟𝑑 𝑥 𝑖𝑛 𝑖𝑡))
- PA algorithm denotes Passive Aggressive algorithm.
- Passive-aggressive algorithms are generally used for large-scale learning.
- It is one of the few 'online-learning algorithms'.
- In online machine learning algorithms, the input data comes in sequential order and the machine learning model is updated step-by-step, as opposed to batch learning, where the entire training dataset is used at once.
- This is very useful in situations where there is a huge amount of data and it is computationally infeasible to train the entire dataset because of the sheer size of the data.
- A very good example of this would be to detect fake news on a social media website like Twitter, where new data is being added every second.
- We took a political dataset, implemented a
TfidfVectorizer, initialized aPassiveAggressiveClassifier, and fit our model. We ended up obtaining an accuracy of 92.11% in magnitude. - Also, with this model, we ended up getting 585 true positives, 582 true negatives, 47 false positives and 53 false negatives.
Real world applications:
- Elections
- Fake job rackets
- Checking the credibility of news links received on social media platforms like Facebook, WhatsApp, Twitter, etc.
- Fake medical news messages
Real world implementation:
- To implement this in real life, we can make a mobile app or a WhatsApp-integrated feature.
- Users would simply enter the link of a news website and be able to verify whether a news website is true or fake.
- In the future, our improved model could be extended to a lot of other data and not just our dataset.