Unsupervised WhatsApp Fake News Detection using Semantic Search

Social media has become the backbone of today’s lifestyle. It has a widespread effect on nearly every walk of life. One of the well-known social media applications WhatsApp Messenger is a free and cross-platform text messaging software that also provides services for sending and receiving multimedia messages. But at the same time in recent years, its easy accessibility has served a way for propagating fake and biased news articles, blogs and messages. Fake news and messages have paved their way for Political polarization, ethnic tensions, unwanted panic and mass hysteria. A solution is proposed that uses Natural Language Processing for analyzing the messages and leverage Transfer Learning Models to detect the authenticity of the information. Claims are filtered from the bulk of forwarded messages disseminated on WhatsApp. The solution comprises of a semantic search mechanism between each claim and associated news sources. The similarity comparison done by the model predicts the truthfulness of the claim.