Interaction Based Credibility Analysis of News on Facebook Using Machine Learning Methodologies

Fake news and its dissemination are growing more popular as social media becomes more pervasive in our daily lives and Facebook users are particularly vulnerable. Numerous studies on detecting fake news have already been published. However, a study focusing on fake news in low resource languages like Bengali propagated through Facebook is rare due to data extraction challenges. In this paper, we work with fake news in the Bengali language and used various supervised machine learning algorithms to classify Facebook posts as fake, real, or satire and found XGBoost to generate the best outputs. Instead of the content, we trained our model with the interaction data of the posts to make it resilient against adversarial attacks. We observed that ensemble methods perform well with such social media metrics.