How the bridging inference links unordered sentences for semantic coherence

With social media becomes increasing popular, volumes of short texts appear in Web, such as Tweets and Micro-blogs. Since these short texts have vast decentralized topics, weak associate relations, large redundancy and abundant noise, how to link this large scale of unordered short text with semantic coherence is a challenge problem. The challenging issues includes: how to represent and measure the semantic coherence state of sentences; how to guild the linking process of short text for semantic coherence via different schemas. To solve the above issues, bridging inference is developed, which simulates the discourse process to narrow semantic gaps between short texts. Bridging inference links unordered short texts by coherence detection and bridging inference schemas. We evaluate our method by measuring semantic coherence in bridging inference process. Experimental results show that bridging inference increases the semantic coherence of unordered short text. The proposed method can be used in short-text origination, e-learning, e-science, web semantic search, and online question-answering system in future works, etc.