Research On User Emotional Polarity Analysis Method In Social Media

With the rapid development of Internet-based social media,social network that reflects and extends the real world has become an inseparable part of people's daily lives.As the main body of social network,people actively participate in social activities,dominate the interaction behavior of the entire social network,and directly affect the social networking environment.Social media not only includes information of human interaction,but also includes structured or unstructured information distributed by users.These massive,irregular,noisy,and hidden social media information brings social network users analysis serious challenges.Opinion leader mining and user emotional polarity classification are the most representative topics in the field of social network user analysis.They have been an important research content for researchers.They are widely used in many fields such as market segmentation,marketing,public opinion analysis and user recommendation.However,based on the traditional centrality measurement which is difficult to apply social network and the inaccuracy caused by the complexity,noise,and irregularity of the text content published by users,this thesis proposes an opinion leader mining algorithm and an algorithm of user emotional polarity classification that combine event sentiment polarity and user's social behavior in events.The main researches of the thesis are as follows:1.An opinion leader mining algorithm is proposed in this thesis,which draws on the core idea of PageRank algorithm and introduces it into social network user influence analysis.The algorithm not only considers some characteristics of social network user's topology,but also integrates social network user's some of the attribute information that reflects their influence.In the aspect of social network topology,the algorithm constructs a weighted user relationship network based on the user's forwarding relationship in the event.At the same time,The events are thematic and involve a large number of users.Building a relational network based on the forwarding behavior between users not only is specific and targeted,but also can significantly reduce the use of social media information data volume,this type of network is dynamic due to different events.At the same time,the algorithm uses user's attribute information that can directly reflect user influence.2.An algorithm of user emotional polarity classification for user social behavior in an event is studied by this thesis.Using textual emotionality measurement,this thesis proposes a sentiment polarity evaluation of event algorithm based on a combination of sentiment lexicon and word frequency statistics.The algorithm not only introduces conjunctions,degree adverbs,emoticons and stop words into basic sentiment lexicon which consists of positive and negative sentiment words,but also analyzes the frequency of words of tweets that are not recognized by the sentiment dictionary,and word frequency and weights is calculated by emotional tweets that have been identified.So we can calculate the emotional polarity of the event.Secondly,user emotional classification algorithm of combining opinion leaders mining and event sentiment polarity is a convex optimization problem in which the opinion leader's attitude value represents the user's sentiment for the topic event and discusses the common attitude of users who participate in the event discussion and have the forwarding behavior.According to the positive and negative attitude value to complete user classification,the algorithm eliminates the use and analysis of a large number of user attribute features,social network topology information,and interaction information.Compared with user classification using text content,the accuracy rate is significantly improved.