Based On The Text Orientation Of Shallow Semantic Analysis

With the arrival of the era of Web2.0, the Internet is no longer an important source for people to acquireinformation, it gradually becomesan important platform for people to express their opinions and feelings, so a lot of subjective textsappeared on the Internet, such as product review information when buying something, and opinion or emotion expressing in social medias. How to extract meaningful units from these subjective texts, analyze and utilize them becomes hotspot.Text orientation analysis produced based on this background. Text orientation analysis has a wide range of application value in the field of product reviews, public opinion analysis and so on.The main tasks of text orientation analysis includeemotional information extraction, emotional information analysis and emotional information application. Emotional information extraction aims to extract meaningful units from subjective texts, transform the unstructured text into structured one which the computer can deal with, emotional information analysis analyze subjective texts from phrase, sentence and text level, and obtain the opinion, emotion and attitude of the authors.Throughparticipatingin the third and fourth Chinese Orientation Analysis Evaluation, this dissertationresearched on the first two tasks of text orientation analysis. The main work and achievements are as follows:Firstly, this dissertationresearched on the opinion target extraction and opinion sentence extraction.Researched different models of opinion targets, and found that nouns with high frequency often appeared in the opinion target. So this dissertationextractedkey words by word frequency as the main basis of opinion target extraction, and introduced word frequency, emotional words, context, chuck and shallow semantic features, put forward anopinion target extraction method based on key words and shallow semantic features.For opinion sentence extraction, this dissertation introduced emotional words, opinion target, interjections, emotional color punctuations, personal pronouns and claim verbs, put forward an opinion sentence extraction method based on multiple features.Secondly, this dissertationconducted the research of orientation analysis from the phrase level, sentence level and text level. Sentiment words can’tindicate the orientation accurately, this dissertation used polarity phrases consisting of sentiment words and modifier words as of the basic unit to express text orientation. Theorientation and strength of opinion target was the orientation and strength of polarity phrases modifying it.The orientation and strength of opinionsentence were based on polarity phrases in the sentence, and considered adversative and progressive conjunction.The orientation and strength of text were based on the sentences,and also considered the influence of text structure, the opinion sentences were divided into three forms:summary opinion sentence, half-summary opinion sentence and ordinary opinion sentence, considering all these aspects, this dissertationput forward a weighted computation method of text orientation based on opinion sentence.In the end, this dissertationcarried on experiments to test the method of sentiment information extraction and analysis. Contracted the methods with others, the experimental results shows that the method put forward in this dissertation was good.