Evaluation of t.rex extracts several signals from the Real-Time web to predict user interest

The real-time Web can provide timely information on important events, and mining it online with stream processing allows yielding maximum benefit from it. In this chapter we proposed an application to online news recommendation. Our goal is to help the user keep up with the torrent of news by providing personalized and timely suggestions. We described T.REX, an online recommender that combines stream and graph mining to harness the information available on Twitter. T.REX creates personalized entity-based user models from the information in the tweet streams coming from the user and his social circle, and further tracks entity popularity in Twitter and news streams. T.REX constantly updates the models in order to continuously provide fresh suggestions. We framed our problem as a click-prediction task and we tested our system on real-world data from Twitter and Yahoo! news. We combined the signals from the two streams by using a learning-to-rank approach with training data extracted from Yahoo! toolbar logs. T.REX is able to predict with good accuracy the news articles clicked by the users and rank them higher on average. Our results show that the real-time Web is a useful resource to build powerful predictors of user interest.