TrustTrace: A Hybrid Mechanism for Fake News Detection in Sinhala Language for Twitter

Abstract

"Throughout history, humans have been referring to “News” as a significant information source for decision-making in their day-to-day lives. With the advancement of technologies and internet services, humans have moved away from traditional news outlets and started consuming online news posted on social media. However, due to the less regulated environment in social media, there is a high chance that people would be publishing fake news to mislead society for personal gain. Over the years, with social media and online news consumption, going up in Sri Lanka, usage of the Sinhala language on social media platforms has also improved with people using the language to share information. Therefore, there is a need for a solution that could safeguard the digital space of Sinhala users. The project TrustTrace proposes a hybrid mechanism that incorporates content-based classification and context-based classification to overcome the dissemination of Fake news in the digital space. The project utilizes Machine learning ensemble methods and a credibility-based scoring mechanism to tackle the problem. A total of 5 Hybrid mechanisms were constructed by combining ensemble learners that predicted with the highest accuracy with the credibility scoring mechanism. Out of the five algorithms, XGBoost combined with sentence embeddings had the highest accuracy levels. The algorithm was able to achieve an accuracy of 84% which goes past one of the similar approaches that has been done for Sinhala FND."