Comment Text Orientation Analysis Technology Research Based On Web

With the rapid development of the Web, the network has become a perfectplatform of exchanging opinions, ideas and showing individual characters. Nowadays,a large number of users express their views and opinions on the platform such asMicro-blog, BBS, shopping nets, with strong emotion tendentiousness. How to mineout these ideas and identify the tendentious of these views more efficiently is the hotspot and focus of the study of natural language areas. Text tendentiousness analysis asthe key technology to solve this problem, mainly is for the user to mine the attitudes,opinions and comments of text of something, and get the attitudes or comments is tobelong to positive or negative opinions. Text tendentiousness analysis has broadapplication space and development prospects in the market forecast analysis, opinionpolls, shopping guide, public comment, film and television evaluation and many otherfields.This paper summarizes the research progress in recent years at home and abroad,analysis the problems the text tendentiousness analysis faced, and puts forward thetrain of thought. In the process of research, of which the key technologies involvedare described in detail, and based on these technologies for comment text tendentiousanalysis work as follows:First of all, this paper studys the extraction of evaluation collocation. It expoundsthe concept of the evaluation collocation, namely the relationship of the evaluationwords and the evaluation objects, specific performance of binary for evaluationobjects, evaluation phrases. Besides, use the maximum entropy method to evaluate thecollocation extraction, In the process of construction maximum entropy template,constructs the evaluation word list, using the synonyms dictionary to evaluation thesynonyms classified, and filling the template with the evaluation word category. Theexperimental results show that the method improves the recognition performance andaccuracy.Second, this paper discusses the construction of the dictionary of polarity indetail. It uses the method of statistics, and machine learning to mine the large-scalecorpus, takes the SogouW from Sogou laboratory, positive/negative evaluationphrases provided by How Net, positive/negative emotional words, the synonymsdictionary and the appraisement righteousness provided by the Chinese praise orblame righteousness usage dictionary as the reference to structure the polaritydictionary. The dictionary constructed includes four parts, such as the basic dictionary,the field polarity word dictionary, the domain attribute words dictionary, the networkdictionary and the modified dictionary. The dictionary is relatively comprehensive,laying the foundation for the tendentiousness analysis.Third, this paper puts forward the polarity calculation formula. It mainlycalculates the polarity by the phrase level, and then calculates the inclination of thesentence level, including the calculation of polarity strength and the judgment ofsentences righteousness. By the polarity dictionary this paper constructed, puttingforward the evaluation value calculation formula of the polarity of the phrase, and constructing the polarity calculation of the sentence level based on the Evaluationphrases as basic unit combined with evaluation objects. This paper uses the corpusproved by the third meeting of the text tendentiousness evaluating, the corpus provedby Tan Songbo, and the corpus grabed by the network to experiment, compared thethree methods in the experiments, The results show that the method this paperproposed were higher than the other two methods in the accuracy, achieve theexpected effect.