Real Time Filtering of Tweets Using Wikipedia Concepts and Google Tri-gram Semantic Relatedness

This paper describes our participation in the mobile notification and email digest tasks in the TREC 2015 Mircoblog track. The tasks are about monitoring Twitter stream and retrieving relevant tweets to users interest profiles. Interest profiles contain the description of a topic that the user is interested in receiving relevant posts in real-time. Our proposed approach extracts Wikipedia concepts for profiles and tweets and applies a corpus-based word semantic relatedness method to assign tweets to their relevant profiles. This approach is also used to determine whether two tweets are semantically similar which in turn prevents the retrieval of redundant tweets.