Analysis And Application Of Sentiment For Network Users

With the rapid development of Internet technology,blog,micro-blog,e-commerce and other social media have emerged constantly.Network users are increasingly inclied to exchange information and share ideas on these network platforms.Such information contains the user's views on various products,news events,organizational groups,and can obtain great commercial value when they get together.Massive network commentary data urgently require automated processing.Sentiment analysis,as an automated analysis approach to network comments,has been extensively studied so far,and plays an important role in many fields,such as enterprise decision-making,public opinion control and information prediction.This thesis explores the emotional tendencies of network users,and focus on the following two aspects: emotional information extraction and sentiment polarity classification.Finally,a sentiment analysis system of network users is designed and realized.In the aspect of emotional information extraction,the thesis puts forward three kinds of methods of emotional information extracting,such as user dictionary,domain related emotional words,and evaluation collocation.Firstly,we propose to use the statistical method to build the user dictionary,which can be used in the process of word segmentation to enhance segmentation effect,and can also be used for the construction of emotional dictionary.Secondly,we propose a co-occurrence graph based method to solve the problem of domain dependence of sentiment polarity.Finally,an algorithm based on syntactic analysis is proposed to effectively extract the evaluation collocation in the text,which includes the method of constructing the graph based lexicon of emotional seed words,and the polarity of the unreached emotion word can be recognized based on the lexicon.In the aspect of sentiment classification,IG-BP classification algorithm is proposed,which mainly includes three processes,feature modeling,feature selection and sentiment classification.We use both product and news data sets to carry out the actual test of the proposed algorithm.The experimental results show that the accuracy rate of the optimal model in the two kinds of data sets is 90% and 87% respectively.Meanwhile the thesis also studies the topic clustering of news text,which aims to aggregate the multi-source news texts of the same topic to support the next sentiment classification.Aiming at this task,the thesis proposes two kinds of text modeling methods,vector space model and probability topic model.And we then improved the K-Means clustering algorithm.The results of clustering experiments show that it is more effective to use probability topic model.In summary,the thesis has made some theoretical innovation in both emotional information extraction and sentiment classification,and has verified the effectiveness of the proposed algorithm through experiments.The implemented prototype system can provide visual display of the results.The research work of this thesis can provide a meaningful reference for the sentiment analysis of products and news corpus.