Research On Opinion Target Extraction For Chinese Microblog

As the most fiery social network platform,the micro-blog text generated every day countless,It covers the news,entertainment,food,goods,military and other fields.Because micro-blog text content-rich and the most able to respond to the current situation and the trend of people's life,so the study of micro-blog's text data is one of the hot spots at presen.In order to find the object that people talk about in large amounts of micro-blog text data,thus has produced this topic of research on opinion target extraction for Chinese microblog.Word segmentation is preprocessing steps of the opinion target extraction for Chinese microblog,The effect of word segmentation directly affect the accuracy of opinion target extraction.This paper around the improvement of opinion target extraction accuracy,do the following research work:(1)Put forward a method of Chinese word segmentation with the good domain adaptive.According to the Conditional Random Field for Chinese word segmentation,the field is hard to adaptive.and thus can't better able to solve the unknown word problem and most of the ambiguity problem.A combination of CRF and domain dictionary is proposed to improve the field adaptability.Put forward a kind of reverse maximum matching algorithm based on Trie tree to calibrate the words segmentation result of Conditional Random Fields and for eliminate ambiguity,we used fixed word collocation,verb dictionary and word probability by the rule of word for-mation.(2)Put forward opinion target extraction method for Chinese microblog with a kind of multiple features of Conditional Random Fields.In order to better extract microblog evaluation object,find the optimal feature template for Conditional Random Fields,this paper to a large number of data were statistical experimental analysis,Analysis of the characteristics and semantic characters,word frequency adjectives location characteristics and the relationship between the evaluation objects,finally formulated the basic characteristics of the part of speech template,semantic role features template,word frequency feature template and describe the word feature template evaluation objectextraction.The multi feature fusion method can effectively improve the accuracy of the opinion target extraction.