Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
424794 | Future Generation Computer Systems | 2016 | 11 Pages |
•Proposing a Markov random field based method for discovering the core semantics of event.•Learning the association relation distribution of event by small scale association relations.•Maximizing the coverage of association relation distribution by the minimum number of short texts.
As social media is opening up such as Twitter and Sina Weibo,1 large volumes of short texts are flooding on the Web. The ocean of short texts dilutes the limited core semantics of event in cyberspace by redundancy, noises and irrelevant content on the web, which make it difficult to discover the core semantics of event. The major challenges include how to efficiently learn the semantic association distribution by small-scale association relations and how to maximize the coverage of the semantic association distribution by the minimum number of redundancy-free short texts. To solve the above issues, we explore a Markov random field based method for discovering the core semantics of event. This method makes semantics collaborative computation for learning association relation distribution and makes information gradient computation for discovering kk redundancy-free texts as the core semantics of event. We evaluate our method by comparing with two state-of-the-art methods on the TAC dataset and the microblog dataset. The results show our method outperforms other methods in extracting core semantics accurately and efficiently. The proposed method can be applied to short text automatic generation, event discovery and summarization for big data analysis.