Identifying Relevant Messages for Social TV

The recent spread of social media, namely microblogging services such as Twitter, has provided a new way to enjoy TV broadcasting. Users of microblogging service can instantly submit messages to express their opinions and feelings while watching TV. However, to use the rich content users generated about TV programs as a valuable source to elicit users´ preferences and understand the programs´ characteristics, we need to identify messages that are relevant to a particular TV program firstly, a task particularly challenging due to some problems such as program title ambiguity and title absence. In this paper we propose a message identifying system which casts the problem of finding TV_related messages as a problem of 0-1 classification. Furthermore, we implement the classification task as a pipeline composed of two stages. We test our method on a dataset composed by approximately 1.5 million messages and experiments show our system achieves both satisfying precision and recall.