Research On Book Domain-specific Sentiment Lexicon Construction And Evaluation

With the continuous development of the information society,the Internet has become the main platform and environment for users to obtain,exchange and release information.With the promotion and development of Web2.0 and crowdsourcing concept,the number and influence of users’ opinions and comments are increasing.The research on user generated content(UGC)has become an important research subject of information science,management science and behavior science research area.As an important sub field of user generated content research,sentiment analysis is of great significance to user behavior analysis,user decision support,public opinion analysis and prediction,information recommendation and evaluation.Sentiment lexicon is an important tool of sentiment analysis and an important object of natural language processing.Book resource is an important tool for the inheritance of human spiritual civilization,an important carrier of knowledge preservation,and at the same time,it has great commercial value in recent years.The sentiment analysis of book resource helps to understand the opinion tendency of users in this field,and provides valuable reference information for each link of the book industry supply chain.In view of the lack of sentiemnt lexicon in the book field,according to the methodology of raising,analyzing,solving and case analysis,this paper uses the book review corpus to construct sentiment lexicon,which contributes to the enrichment of the theory of sentiment lexicon construction in the field and provides reference for the related practical work.The first chapter is the introduction.This research clarifies the theory background and significance,expounds the importance and necessity of the construction of sentiment lexicon in the field of books,puts forward the theoretical and practical value of this work,and emphasizes the theoretical and practical significance of this paper.Based on the analysis of the current research situation at home and abroad,this paper introduces the research object,research framework and content,as well as the research methods and innovation of this paper.The second chapter is the theoretical basis.Theoretical basis of this paper includes machine learning theory,text feature analysis theory,user behavior theory.Supervised learning in machine learning theory provides a method for the effect evaluation of lexicon.Text feature analysis theory provides a method for the extraction,association and scoring of emotional words and attribute words,while user behavior theory provides a perspective and research model for better analysis and understanding of user behavior.The third chapter is about the analysis of sentiment lexicon characteristics in the field of books.The analysis of user behavior characteristics and corpus structure is the basis of dictionary construction.The user comment data of Douban reading,the largest book tagging website in China,is collected to analyze the part of speech,frequency and relevance distribution of characteristic words.The expression habits and behavior patterns of users are further analyzed through the characteristics of words,which is the basis of lexicon construction in the following article.The fourth chapter,based on the ultra short review of the book field sentiment word recognition and strength judgment part.To solve the problem that word segmentation tools have a great influence on the final sentiment lexicon construction results,we bypass the word segmentation stage,build an emotion dictionary based on the user’s ultra short comments,take the user’s evaluation marks as the emotion tendency and intensity,use the frequency of words in the positive and negative marks to replace the previous point mutual information calculation method based on seed words,and use crowdsourcing wisdom as the sentiment intensity after determining the tendency Assign value,construct basic sentiment lexicon.The fifth chapter is the attribute word recognition and association part,which integrates deep learning and syntactic rules.Attribute word recognition is an important related work in the construction of sentiment lexicon,and it is also an important basis for judging the ability of specific reference in the field of sentiment words.This part integrates word2 vec model and important syntactic structure in deep learning,uses the seed attribute words of manual annotation for iterative recognition,and then uses the relationship between attribute words and sentiment words to rise a method to calculate the domain characteristics of sentiment words by link analysis.Chapter six,the evaluation of lexicon effect,taking the evaluation of book review quality as an example.The ability of sentiment lexicon in the book field needs to rely on its application effect in specific scenarios for evaluation.In this chapter,in addition to the accuracy rate and recall rate of off-line evaluation indexes,the user generated content quality evaluation is taken as the application scenario.Compared with other general emotion dictionaries,the effectiveness and availability of sentiment lexicon in the book field in this paper are proved.Chapter seven,conclusion and prospect.This paper combs,summarizes and summarizes the whole work,points out the advantages and contributions of this construction method and sentiment lexicon,analyzes the limitations and shortcomings of the research,and looks forward to the future extension research.Through the content of the above seven chapters,the paper expounds the user behavior characteristics,sentiment lexicon construction,attribute word extraction and association,sentiment lexicon ability evaluation and other contents in the book field in a structured and hierarchical way,take the advantage of qualitative and quantitative therory,to provide reference for the solution of related problems.