Fake News Detection

This repository contains code for building a fake news detector. This project was developed during Data Science for All Women's Summit 2020.

Problem statement

• Fake news is a particularly important problem nowadays as many people rely on social media as their primary news source.¹
• Impact of fake news
  • Social: Fake news spread faster and further than real news.²
  • Economic: Fake news affects stock market. For example, a hacked AP tweet in 2013 led to a 130 billion dollar loss in market cap.³
  • Business: How social media platforms handle fake news affects customer trust.
• Our solution is to develop a model for classifying news articles as real or fake.

Data

• Model training: Fake and real news dataset (Kaggle).
• External validation: Fake news dataset from 2016 (Kaggle).

Analysis

• Exploratory data analysis & text preprocessing
• Baseline model 1: doc2vec embedding & logistic regression
• Baseline model 2: Recurrent Neural Network
• Advanced model: BERT & transfer learning
• Model interpretability: LIME
• External validation

Results

Please find our results in our project report or click on the following image to view our slides.

Authors

Iris Yoon
Rabiya Noori
Jerri Zhang
Renee G. Reynolds
Hannah Mei

References

1: Americans Who Mainly Get Their News on Social Media Are Less Engaged, Less Knowledgeable
2: A survey of fake news
3: False Rumor of Explosion at White House Causes Stocks to Briefly Plunge