Text Similarity-Based approach to detect Sinhala Language Fake News in Social Media: An approach using Hybrid Features

Abstract

In this rapidly evolving digital age, societies rely heavily upon social media to share news publicly due to the speed of dissemination. With billions of users, the sharing of a news item only takes a few minutes to represent diverged views, with malicious or misleading content, to go viral. In 2018, Sri Lanka experienced anti-Muslim riots, and in 2019, racist uprisings initiated fake news in social media mainly in Sinhala language. Considering the massive number of Sinhala language posts shared at present and the deficiency of research work in Sinhala fake news detection, an automatic fake news detection technique is proposed in this research that can help to identify fake news published in the Sinhala language which circulates on social media sites. Approaches to detect fake news depend heavily upon features inherent to either the explicit or implicit features of user account and text content-based features of the post, or any hybrid set of above features. Based on the literature, social media users mainly consider verifiability to identify fake news content. Therefore, the hybrid methodology proposed in this research workmainly focused on the checking and verifying whether the news text content appears on the credible sources. The authenticity features of the user account that used to obtain the news content were evaluated in the rule-based points allocation schema. An accuracy of 78%was gained in predicting fake news with this Rule-Based implementation.