Research On Multi-feature Sentiment Lexicon In Sentiment Analysis

With the rapid development of China's Internet industry,how to extract effective information from the growing mass of data has become a hot research topic.As an important application direction of natural language processing,text sentiment analysis can be used to analyze,process and judge the emotional tendency of text.It is also widely used in public opinion monitoring,credit evaluation,network marketing,product and service optimization and other fields.Sentiment lexicon is the basis and important tool of text emotional analysis.Perfect and accurate sentiment lexicon can effectively improve the effect of text emotional analysis.Most of the existing general sentiment lexicons are composed of emotional words and emotional polarity values,and their emotional tendency is usually set to a fixed value.This kind of sentiment lexicon has strong subjectivity,does not consider the different emotional tendencies that emotional words may show in different contexts,and has a certain one-sidedness.The length of sentences,the frequency and location of words,and the relationship between words contain a large amount of text feature information,which is the key to the change of emotional intensity of words.In this paper,the text feature information of words is added to the construction of sentiment lexicon in order to find the relationship between the feature information of words and emotional tendencies in different text contexts,so as to make the sentiment lexicon better applicable to all fields.This paper summarizes more than 30 kinds of common text features and classifies them according to their categories and ranks,in order to find out the features that have an impact on the emotional value of words in each category.After a lot of research on the relationship between the textual features of words and emotional tendencies,this paper proposes a feature-based approach to constructing sentiment lexicons.The basic idea of the algorithm is to select text features,count the eigenvalues and emotional values of words in the corpus each time they appear,and construct the list of eigenvalues-emotional values of words.Clustering method is used to analyze the emotional value of words in different eigenvalue ranges,and find the relationship between the emotional value and the eigenvalue.The valid emotional value and corresponding eigenvalue range are stored in the sentiment lexicon by judging the admission conditions.Different features of words contain and reflect different aspects of their textual information.A single feature can not fully express the emotional information of words in the context,but its actual effect has limitations.In order to solve the problem that a single feature contains limited emotional information of text,based on the construction of feature-based sentiment lexicon,this paper also proposes a Multi-feature-based sentiment lexicon construction method.The main idea of this algorithm is to integrate the advantages of each feature and improve the effect of multi-feature dictionary by comparing the results of single-feature affective dictionary and combining the best features of each feature.After grouping and comparing experiments,this paper compares the single feature affective dictionary constructed based on each feature,and combines the actual features to construct a multi-feature affective dictionary.The experimental results show that the classification effect of multi-feature affective dictionary is better than that of nonfeature general affective dictionary and single-feature affective dictionary.This proves the validity of the new feature-based sentiment lexicon construction method proposed in this paper.