Research On Opinion Leader Mining Technology For Symbolic Social Networks

Opinion leaders in social networks refer to users with high prestige.They are active on social network platforms,expressing opinions on social focus and guiding public opinion.Identifying opinion leaders accurately enables the government to grasp the trend of public opinion and control the public opinion in time.At present,opinion leader mining methods are mostly based on user attribute or network structure.These methods only consider user attribute or user behavior,but ignore user relationship.Therefore,the sentiment analysis technology of comment text is studied to get more precise characteristics on user relationship.At the same time,combined with user behavior and user attribute,an opinion leader mining method based on multi-feature fusion is proposed.The specific work is as follows:(1)A sentiment analysis method of comment text based on Parallel Hybrid Neural Network is proposed.On the one hand,for the issue that semantic features of comment text are hard to fully extract,Convolutional Neural Network is used to extract local features of text vector,and Attention-Bidirectional Long Short Term Memory is used to extract global features related to text context.On the other hand,for the issue that distributed word vector lacks attention to sentiment information and classification contribution,sentiment weight is integrated into TF-IDF algorithm,and weighted word vector is generated.Experiments have been conducted on the data set of comment texts from Sina Weibo.The experimental results show that the macro_F1 value of the proposed method is 90.75%,which proves that the sentiment analysis method proposed in this thesis can accurately mine the sentiment tendency of comment text.(2)An opinion leader mining method based on multi-feature fusion is proposed.First,signed network is used as the research tool to assign the relationship symbols that represent support or opposition among users.In this way,user relationship in the real society can be more objectively reflected.Second,the coefficient of variation method is used to distribute the weights of user attribute.By determining the weight of each attribute more objectively,human errors can be minimized.Third,the method integrates the characteristics of user behavior,user relationship and user attribute,which can comprehensively measure the ability of communication,degree of support and level of activity of opinion leaders.Experiments have been conducted on the data set of hot topics from Sina Weibo.The experimental results show that the coverage rate,core rate and user support degree of the proposed method are better than other methods,which proves that the opinion leader mining method proposed in this thesis can accurately mine the opinion leaders in social networks.