Research On User Influence And Short Text Emotion Toward Weibo Topics

Sina Weibo has rapidly developed into an important media platform with a large number of ordinary users and well-known users in various fields.Combining user influence and micro-blog text sentiment classification to analyze the emotions of ordinary users and influential users in micro-blog topics,it plays an important supporting role for each department and industry to take the next step.However,the existing researches mainly have the following problems:(1)In terms of user influence,Short microblog makes LDA topic model mining ineffective,the dynamics of user interaction time and the relevance between users and interactive topics are not considered enough;(2)In the micro-blog emotion classification,the short colloquialization of Weibo leads to poor classification effect,insufficient emotional relationship between the introduction of Weibo,and strong dependence of classification performance on the scale of training set.This paper conducts research on the above issues,as follows:(1)In order to make LDA topic model better analyze microblog topics,in order to mine the topic areas where users are good at,a method of calculating user influence based on microblog topics is proposed.In view of the irregularity of microblog's short size,this paper uses microblog social relations to improve LDA topic model to analyze microblog topics.In view of the lack of dynamic consideration of user interaction time,the exponential decay function is used to simulate the dynamic time of user forwarding time,and it is introduced into the calculation of user forwarding influence and user's own quality,aiming at ignoring the relationship between users and topics,solved by calculating the degree of association between users and interactive topics.The experimental results show that the method can effectively identify users who can continuously produce high influence in the topic domain,and the calculation performance is better.(2)In order to analyze the emotional polarity of users under the topic of Weibo,a microblog emotion classification method based on topic relationship and user influence is proposed.Aiming at the problem of insufficient validity of microblog emotional relationship,the user's topic context,explicit consent relationship,and implicit user influence relationship are used to establish the emotional relationship between microblogs to solve the problem.In view of the problem of poor classification caused by short microblogging,the emotional polarity of emotional ambiguity is determined by introducing the network dictionary and establishing emotional relationship between emotions and microblogs.The dependence on classification performance on large-scale training sets is strong.The problem is to establish a semi-supervised sentiment classification method for the emotional relationship between labeled and unlabeled samples,which can alleviate the dependence of classification performance on the training set to some extent.The experimental results show that the classification performance of this method is better than the traditional microblog emotion classification method.(3)In order to effectively eliminate the interference of network water interference,provide accurate and reliable stock sentiment information,and realize the stock investor sentiment orientation analysis system for stock topics.The system applies user influence and Weibo sentiment to the stock field to identify high-influence investors and ordinary investors in massive stock investors and to classify their corresponding sentimental tendencies.The main functions of the system are investor classification,investor sentiment orientation classification,and relevance display.Using real stock microblog and stock historical data,it is verified that the system can alleviate the lack of stock microblog corpus to some extent,and empirically analyzes that the system can effectively eliminate the interference of stocks online supporters by combining the two,which has certain significance for the research of stock market.