High quality information extraction and query-oriented summarization for automatic query-reply in social network. (February 2016)

• Record Type: Journal Article
• Authors: Peng, Min; Gao, Binlong; Zhu, Jiahui; Huang, Jiajia; Yuan, Mengting; Li, Fei
• Is Part Of: Expert systems with applications. Volume 44(2016)
• Page Start: 92
• Page End: 101
• Publication Date: 2016-02
• Subjects: Query-reply -- Social network -- Information extraction -- Summarization; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2015.08.056
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: We use information extraction method to get useful messages of social network. We use summarization method to reply the query in social network. We pay more attention to reducing noise and eliminating redundancy. Our method performs well in both automatic evaluation and manual evaluation. Abstract: In this paper, we propose a new method for automatic query-reply in social network. Information extraction and query-oriented summarization method are applied here to reply people's query. There are few effective and commonly used methods on filtering the redundancy and noise of the raw data, which results in the poor quality of the reply. Due to the characteristics of social network messages, we pay more attention to reducing the noise and eliminating the redundancy of the messages to ensure the quality of the final reply. First, we propose an information extraction method to extract high quality information from social network messages, which is based on time-frequency transformation. Second, query-oriented text summarization is implemented to generate a brief and concise summary as the final reply, which is based on the scoring, ranking and selection of sentences of high quality social network messages produced by previous step. Experimental results show that the research is effective in filtering the redundancy and noise of social network messages, the final query-reply results outperform other commonly used methods' results in both automatic evaluation and manual evaluation. Through our approach, noise and redundancy of social network messages are effectively filtered. Certainly, our method improves the quality of the reply for people's query.