Research on Extension LexRank in Summarization for Opinionated Texts

Along with the boom of information, it is an important task to summarize the valuable information of opinionated texts. Firstly, we model the opinionated text by TAM and get the topic and aspect attribute of every sentence. Secondly, we successively used the basic LexRank, Comparative LexRank, Topic-sensitive tf*idf LexRank and Topic-sensitive tf*idf & Comparative LexRank to generate summary. Experimental results show that the best summary on precision can be gotten by the topic-sensitive tf*idf LexRank and the best result on recall and F-measure can be gotten by the topic-sensitive tf*idf & comparative LexRank.