Topic and Opinion Classification Based Information Credibility Analysis on Twitter

At the Great Eastern Japan Earthquake in 2011, a huge amount of information about the disaster were exchanged on Twitter. On the other hand, various false information and rumor were also spread on Twitter. Therefore, it is required that people easily check information credibility. In this paper, we propose a method for automatically assessing the credibility of information based on the topic and opinion classifications. We assess the credibility of information by calculating the ratio of same opinions to all opinions about a topic. For identifying which topic is mentioned in a tweet, our method uses topic models generated by Latent Dirichlet Allocation. For identifying whether an opinion of a tweet is positive or negative, our method performs sentiment analysis using a semantic orientation dictionary. We performed our experiments on 2960 tweets and show more than 0.6 in kappa statistics between our method and human scorers.