Discriminant Model Of Influencers Based On The Relative Emotion

Based on the power-law distribution of social networks centrality,most individuals form the long-tail of the network,and only a few individuals become influencers.In the field of social networks,the discrimination of influencers is a research hotspot.With the rapid development of new media technology and the increasing attention paid by the state to the supervision of network public opinion,how to effectively identify influencers has attracted more and more attention from scholars in sociology,computer science and complex system dynamics etc.At present,the measurement of user influence mainly focus on the diffusion parameters of information in the network,such as the speed,range and depth of information propagation.This measurement essentially characterizes the size of information propagation capability,ignoring its critical polar state.In real social networks,the polarity of user influence is a universally existing basic attribution.Simply from the perspective of binary classification,the polarity of user influence is divided into positive and negative influence.Influences with different sizes and polarities are mapped to user attributes to form different influencers.From the respective of sentimental polarity,this article first designs the comprehensive influence ranking(CoInf-Rank)algorithm for influence to effectively discover the set of influencers.Then,this article proposes the influencers classification model based on relative sentimental polarity,which divides influencers into three categories: opinion leaders with positive sentimental influence,trolls that arouse negative sentiment,and controversial figures who trigger fierce debate.Finally,the effectiveness of the model is verified by experiments and empirical analysis.The main contents of this article includes:(1)Design the CoInf-Rank algorithm that integrates multi-centrality and behavioral characteristics.Firstly,construct a social network graph based on user comment relationship.Then,use the multi-centrality indicator to measure the influence of user nodes in the social network graph.Finally,incorporate the quantitative behavioral characteristics indicator to mine topics users that in key positions,with a wide range of influence and strong influence.(2)Construct a topic impact user classification model based on relative sentimental polarity.First of all,using the sequence labeling model(BiLSTM-CRF)trained by the manual labeling data sets to extract emotional elements automatically.Then the proposed Emotional Matching and Emotional Transforming algorithms are leveraged to calculate the relative sentimental polarity value between users.Finally,according to relative sentimental polarity of users,the influencers are divided into opinion leaders,trolls and controversial figures.(3)Design experiments to demonstrate the feasibility of the proposed algorithms and models.The CoInf-Rank algorithm is applied on ranking the influence of topical users with experiment results presenting that the CoInf-Rank algorithm can identify influencers more comprehensively.Compare the relative sentimental polarity discrimination model with the benchmark model,and experiment results show that the relative sentimental polarity discrimination model has higher accuracy.Meanwhile,empirical analysis confirms that opinion leaders,trolls and controversial figures have different roles in social networks,and it is necessary to further division.