Research And Implementation Of User Influence Analysis Model Based On Topic

With the widespread application of social networks,people use social networks such as Weibo,Twitter and Facebook to learn about current affairs,express their opinions and exchange information,thus achieving the purpose of social contact.As a social platform,social network is also a key platform for brands to market their products online.Businesses place advertisements through high-influential users,whose fans then retweet and spread the advertisements to their followers,ultimately achieving the purpose of recommending products.How to find users with strong ability to promote specific ads is the challenge for the researchers.To this end,this paper first proposes a topic-specific social network influence assessment model,which combines a semi-supervised topic model and PageRank by introducing topic-specific seed words to calculate the influence of users under different topics,thus extracting high-influential users of the target topic;and optimizes the model by introducing topic interactions and improvements on user similarity assessment.The model was validated on the Weibo dataset.Then,to address the problem of malicious account detection,a user credibility assessment model based on hierarchical attention and graph neural network is proposed,which models the semantic features of users through the text content published and retweeted by users,obtains the interaction between users through the user retweeting network,analyzes the features of users in social networks,and finally uses the credibility rating to filter out high-quality users with a low proportion of published rumors;the model is improved by introducing a multi-layer attention mechanism to extract key words and key microblogs.Finally,user behavior is analyzed by a predictive model of user retweeting behavior based on the topic model and convolutional network,and user interest features are extracted from the distribution of user topic preferences,and structural features are used to calculate the influence of neighbor users,so as to filter out the neighbor networks with greater influence on the target users and construct local network features.Combining user topic preference,local network features and credibility features,we model the multi-dimensional user behavior to predict the retweeting behavior.The model achieves better results in all metrics compared to the existing DeepInf,a retweeting prediction model based on structural features of user nodes and randomly sampled neighbor network features.This paper evaluates and analyzes users from multiple perspectives,and applies influence assessment and credibility assessment to retweeting behavior prediction which verifies the effectiveness of each model.It has important practical implications for applications such as advertising and online marketing.