关键词: 中文博客, 情感分析, CRFs

Abstract: In the field of Chinese blog sentiment analysis，previous researchers put most energy on the polarity analysis of word，but not all the word analyzed is relative with the topic，and word-level granularity for sentiment analysis is too small.We try to use sentiment sentences，a sentence-level model，for sentiment analysis.In this paper，it only focuses on topic sentiment sentences auto-extraction.In order to extracting topic sentiment sentences，it designs a novel Bi-segment method to extract the main topic words，and uses TFIDF to extract more topic words.With these words，it recombines original sentences，which contain the topic words.So as long as topic sentiment sentences exist，they must in the set of recombined sentences.Then，based on the analysis of Chinese blogs，it converts the problem of extraction into Chinese chunking by CRFs and has a good performance in extraction experiment.

Key words: Chinese blogs, sentiment analysis, Conditional Random Fields（CRFs）