Blazers

Inspiration

This is a project to combat fake news and misinformation spread especially in the Finance Sector. The spread of misinformation and fake news is ranked as one of the world's top global risks by the World Economic Forum. Fake news is estimated to cost the global economy around $78 billion per year. In a single incident, fake news caused a $300 billion loss in the stock market. Based on an analysis of previous cases involving fake news causing damage to global stock markets, we estimate that a potential loss of up to 0.05% of stock market value is at risk due to fake news. As a direct result of fake news, this amounts to a $39 billion annual loss. Not only, are the costs direct but some of the costs are indirect also. These indirect costs may have an impact on quality of life, increase fear, or change behaviour. Hence, we decide to create the API to tackle this issue. Misinformation and the spread of fake news have caused adverse effects on individuals, communities, and society in recent years, making the problem of fake news detection very relevant in current times. There are several existing solutions to this classification problem including traditional Machine Learning and Deep Learning models which are trained purely on intrinsic features extracted from the news article. However, none of the above methods utilises extrinsic features like the user propagation graph of the article to determine its validity. Our API uses the innovative Graph Neural Network (GNN) approach to tackle the problem of fake news.

Our Project

Our web-based application takes a URL of a news article and determines its veracity using a two-pronged approach - incorporating user data as well as Twitter propagation graphs - and cutting-edge technology - Graphical Convolutional Networks (GCN).

Challenges

Twitter-API has a 15-call limit every 15 minutes. To make our application more efficient and user-friendly, we built custom web-scraping algorithms for data wrangling and scheduling. Developing a strategy for encoding users' tweet histories using word embeddings and the feature vector of their respective user profiles.