Information credibility: A probabilistic graphical model for identifying credible influenza posts on social media

Abstract

Social media is an important data source to compliment traditional epidemic surveillance. However, misinformation in social media hinders the exploitation of valuable information. Analysis of information credibility has drawn much attention of academia in recent years. In this paper, we focus on analyzing the credibility of influenza posts published on Sina Weibo. We propose a semi-supervised probabilistic graphical model to jointly learn the interactions between user trustworthiness, content reliability, and post credibility. To test the performance of the approach, we apply it to identify credible influenza posts published from May 2013 to June 2014 on Sina Weibo. Random Forests and the Bayesian Network are used as baselines for evaluation. The results show that our approach performs effectively with the highest average accuracy of 71.7%, f-measure 51%. Our proposed framework significantly outperformed the baselines in detecting credible influenza posts on Sina Weibo.