A Study On Automatic Summarization For Chinese Opinion Texts

With the rapid development of web in China, we almost find theopinion expressed everywhere. Although opinion will be very helpful, weneed to spend a lot of time to find and check it since the abundantinformation. So there is great practical application value to summary thosetext containing opinion. Different from the traditional text summarization,the target of summarizing opinion text is to extract the opinion text fromonline text, and generalize on the redundant opinions information.We propose an approach that uses the typed dependency to extract thefeatures and opinion pairs in review and uses the classification algorithm forclassifying the correlation between the feature and opinion to improve theprecision of extraction.Then, we adopt the sentiment lexicon to classify the sentiment ofreview. We convert the summarization generation to the p-median problem,and define the most critical cost function of p-median problem from twoangles, that is from the contained information in the review and from theprobability distribution of opinion strength. And we use an approximatesolution of a p-median problem to solve the problem of summarizationgeneration.Finally, we conduct the experiment of extraction of features andopinion pairs, correlation of the pairs, the sentiment to the features of thereview and the summarization generation, as well as compare theexperimental results with the domestic and foreign related research results,which shows that the proposed approach in summarizing Chinese opiniontext is practical and feasible.