Research On The Self-construction Method Of Emotional Dictionary Oriented To Specific Topics In Social Networks

Sentiment analysis is a hot topic in the analysis of big data analysis in current social networks.The current social network sentiment analysis can be divided into two research directions,one is sentiment analysis based on sentiment dictionary,and the other is sentiment analysis without sentiment dictionary.Sentiment analysis based on sentiment dictionary can effectively improve the accuracy of sentiment analysis by continuously optimizing sentiment dictionary.However,sentiment analysis without sentiment dictionary often has relatively low accuracy and high analysis cost.In the field of sentiment analysis based on sentiment dictionary,sentiment dictionary is the cornerstone of sentiment analysis.Therefore,how to construct an efficient sentiment dictionary is an important research direction of sentiment analysis.The traditional sentiment analysis uses universal sentiment dictionary,authoritative universal sentiment dictionary,such as NTUSD at Taiwan University and HowNet sentimental dictionary.However,these dictionarys are relatively simple,and they divide sentiment into positive and negative or positive,negative and neutral.These traditional classifications cannot fully meet the needs of the sentiment analysis under the current changing social network topics and human complex sentiment dimensions.In addition,in different topics of social networks,the problem of "word polysemy" of sentiment words also brings a challenge for universal sentiment dictionary which is used for sentiment analysis.Aiming at some problems of traditional sentiment dictionary,we propose a selfconstruction method to construct sentiment dictionary in a specific topic based on spectral clustering(SDSC).This method addresses the problem of polysemy in different topics from the source of data which be used to construct sentiment dictionaries.SDSC model includes FT model(Filter Text model),CRM model(Construct sentiment Relationship graph model),SC model(Spectral Clustering model)and related calculation methods and theories.In this regard,the main work of this paper is as follow:1)We propose the FT model,in the FT model,we filtered and obtained useful comments as a dataset to construct the sentiment dictionary by calculating the number of comment's forwarding,the number of points,the number of comments,and the increment of these interactions per unit time.2)We propose the CRM model,in the CRM model,the sentiment similarity between sentiment words is used as the weight of the edge,and sentiment words are used as nodes to construct a weighted undirected graph of sentiment relationships,our sentiment similarity includes basic sentiment similarity,topical sentiment similarity,and synonym sentiment similarity.3)We propose the SC model,in the SC model,we divide the undirected graph of sentiment relations into three subgraphs and five subgraphs by spectral clustering algorithm,and identify the central words of each subgraph through the correlation calculation method of the central word detection,and finally output the sentiment dictionary.4)Combining the above three models,we propose the SDSC model which based on spectral clustering.5)The experiment shows that the sentiment dictionary output by the SDSC model has a high accuracy.At the same time,the model is simple,flexible,and efficient.It can build a domain sentiment dictionary better,solve the problem of polysemy in different topics,and improve the accuracy of sentiment analysis.