Social media has gained impressive popularity all around the world in the last decade. Social networks such as Twitter, Facebook, LinkedIn, and Instagram have acquired their user’s attraction by maintaining their identity with very similar features. With the popularity of these platforms, now a day most of the users tend to rely on the information published on social media. Therefore, the credibility of social media information is playing a major role in the present cyberspace. As an example, the Twitter platform is handling 500 million tweets per day. Most of the twitter messages are truthful, but the twitter platform is also used to spread rumors and misinformation. Truthfulness or reliability is depending on the source's credibility. Twitter profiles can be identified as the information source on the twitter platform. In this paper, a user reputationbased prediction method is proposed to analyze the twitter source credibility. The proposed solution is mainly based on the k-means clustering model. Another two models namely, news category analysis and sentiment analysis are deployed to generate novel features for the clustering method. The objective of this paper is to introduce a credibility rating method to visualize the user credibility of twitter user profiles. So that followers can have an understanding about the trustworthiness of the information published on that profile. Producing the agreement score for a specific twitter user is one of a novel experiment in this research. Achieved accuracy by the system is 0.68 according to the evaluations conducted.

Abstract

Religion is one’s relation to what he or she regards as holy, sacred, spiritual, or worthy of especial reverence. Religious extremism is the advocacy of extreme measures over a religion whereas religious extremists are even willing to murder as they provide sanctions for violence in the service of God. Sri Lanka has a tragic history of religious and ethnic extremism and the Easter Sunday attack coordinated by a radical Islamic group that killed over 300 and injured another several hundred can be identified as the recent climax of these events. In this modern information age, it is evident that these radical extremist groups utilize social media for spreading their extreme ideologies due to its free and unregulated nature. If there were a mechanism to even slightly identify the possibility of tragic incidents like Easter Sunday bombing, the 300 souls who had to sacrifice their lives for an unreasonable cause would be still alive happily. In this research, we propose a predictive methodology for identifying any upcoming religious extremism-based threats in Sri Lanka using social media intelligence. We aim to specifically address Sri Lanka’s multi-lingual culture by analyzing all the bilingual social media posts in Sinhala and Tamil languages. A hybrid sentiment analysis methodology consisting of a Machine Learning model and a sentiment lexicon was trained on carefully chosen labelled social media text data and each text was classified as either religious-extreme or not, using Naïve Bayes, SVM, and Random Forest algorithms. When comparing their results, we were able to achieve the best results with the Naïve Bayes algorithm resulting in an accuracy of 81% for Sinhala tweets while Random Forest algorithm resulted in an accuracy of 73% for Tamil tweets proving that social media intelligence can be used to predict religious extremism-based threats.