A Collision of Beliefs: Investigating Linguistic Features for Religious Conflicts Identification on Tumblr

Research shows that with the unexpected emergence of religion and faith, identifying religious conflicts within society has become an important problem for the government and law enforcement agencies. Many social science researchers and domain experts conduct manual surveys on offline and online bases for finding such conflicts. On the other hand, it is seen that people use social media websites for sharing their religious opinions, sentiments and beliefs. We create a hypothesis that social media websites are a rich source of information for mining these beliefs and automatically identifying the religious conflicts among users which overcomes the gaps of offline studies. In this paper, we address the challenge of ambiguity and multilingual scripts in social media posts and distinguish them into various religious sentiments of users. In order to evaluate our hypothesis, we conduct our study on Tumblr- the second most popular online micro-blogging service. We create a dataset of all Tumblr posts (published since 2007) consisting of several tags commonly used in religion based posts and make it publicly available for benchmarking and comparison. We investigate the efficiency of natural language based features for identifying the Tumblr posts that discuss about a religion and belong to one of the nine categories of users’ sentiments. For example, disagreement, defensive, annoyed and disappointment. We manually analyze these posts and our result shows the proposed features are discriminatory and support our hypothesis. Furthermore, our results reveal that despite the subjectivity in Tumblr posts, it is technically challenging to mine the religious sentiments of bloggers.