Opinion Mining Based On Network Text

At present, information on the internet is increasing rapidly. On the internet, peoplecan get more information that they have a interest than ever. Meanwhile, the informationof reviews occupies a very large proportion. Based on the characteristics of the internet,those information is very complex, those useful information and invalid information isoften mixed together. It is often very difficult for people to obtain valuable informationfrom reviews about certain persons, events, media, products in a short time. Therefore,data mining from the reviews such as useful information extraction has become a veryimportant task today.In this paper, we mainly focus on the text filtering for useful reviews, the subject ofreviews and sentiment analysis. In the work about text filtering, this paper implements thefiltering of unrelated words in the text of reviews, and the filtering of unrelated reviewsthat have a low correlation with the subject text. Besides this, this paper implements theidentification of words that are parts of subject，the judgment of sentiment tendency foreach specific set of reviews and the overall reviews.For review text filtering, this study firstly use a sophisticated segmentation tool for theword boundaries and speech identification on the Chinese text. Then it achieve the uselessword filtering combined with a dictionary of useless words. After this step, it implementsthe filtering of those reviews that are not related with the subject based on the text distanceanalysis. In extraction for the subject of reviews, this paper implements the identificationof reviews’ subject based on the study of Chinese nouns and noun phrases, grammar rulesof words’ combination, training sets of review text and associated words that appearfrequently. In the step for the sentiment tendency analysis, this paper consider the problemfrom bottom to top and implements an unsupervised learning approach for the vocabularyexpansion of words that show sentiment tendency. Based on this, it achieve identificationof the sentiment tendency in review text.