Abstract
The increased usage of Twitter as a medium for reporting news and sharing information between people has caught the attention of researchers from different disciplines. One of the research directions is the analysis of online information from the perspective of its credibility. This paper aims to assess and analyze the credibility of tweets in Arabic language. In order to achieve the stated goal, first we employ the idea of crowdsourcing where users can explicitly express their opinions about credibility of a set of tweets. This information coupled with the data about tweets' features enable us to investigate which features may indicate the credibility level of a tweet, e.g. tweet with attached image and was authored by a person who posts a lot of tweets will be, with high probability, a credible tweet. We distinguish three main groups of features: authority and topical expertise (of the source), data quality (of the content), and popularity (of the content and the source). We argue that content data quality factor based on content linguistic features in addition to source authority is more important than content popularity in identifying credible messages. In addition to this, we identified three experts who also rated the credibility of tweets and based on that we investigate the level of agreement between experts and the crowd, and we identify which expert represents the crowd in the best way. This can allow us to select the most representative expert when it is needed. This study is a pilot of a large study that aims at predicting credibility of Arabic Twitter messages using machine learning approaches.