Domain-Specific Automatic Summarization Strategy

Automatic summarization is a traditional research field of text mining. In this paper, two domain-specific automatic summarization techniques; mobile summarization, opinion summarization which includes opinion-holder based opinion summarization and feature-based opinion summarization are designed.Mobile oriented automatic summarization is restricted to summary length due to the smaller screens. In this paper, an improved String-edit Distance-based mobile summarization technique is designed to create the summary displayed on the mobile terminal. Considering some web pages are structured with subtitles, hierarchical summarization is applied to them in order to improve the coverage of the summarization.Opinion-holder-based automatic summarization is to organize the opinion by the opinion-holder. In this paper, a Chunk-CRF model is constructed to divide the opinionated sentence into particular chunks with the aim of effectively identifying the opinion sources; at the same time, in multi-Opinion-Holder cases, syntactic analysis is made use of before applying CRF model. On this basis, opinion summarization and polarity analysis are given.Feature-based opinion summarization is to summarize a product by several features. In this paper, a Condition Random Field (CRF) model is trained in order to assist the comparative relations and feature extraction. On this basis, feature merge and polarity analysis are made use of to create an optimized feature-based opinion summarization and visualization result.Experiments showed the summary created by the mobile summarization in this paper does well in conciseness, readability and coverage, moreover, the effectiveness hierarchical summarization is proved by a Q&A evaluation; the ChunkCRF-based method which identified opinion holders with precision over 80% could assist the opinion-holder-based opinion summarization well; Chinese Comparative Relation Extraction contributes to the solution of product feature extraction and polarity analysis, which lays solid foundation for the feature-based opinion summarization.