Identifying News from Tweets

Informal genres such as tweets provide large quantities of data in real time, which can be exploited to obtain, through ranking and classification, a succinct summary of the events that occurred. Previous work on tweet ranking and classification mainly focused on salience and social network features or rely on web documents such as online news articles. In this paper, we exploit language independent journalism and content based features to identify news from tweets. We propose a novel newsworthiness classifier trained through active learning and investigate human assessment and automatic methods to encode it on both the tweet and trending topic levels. Our findings show that content and journalism based features proved to be effective for ranking and classifying content on Twitter.