Research On Topic-oriented Social Network Influence Maximization

With the continuous socialization of information networks,social networks are gradually being integrated into our lives.Just as weibo makes topics,zhihu shares knowledge,and postbar creates fan culture,social networks have become a kind of lifestyle of more and more people and the way to access information.How to excavate key users from social networks for effective dissemination to maximize its impact is of great significance in the field of public opinion guidance and virus marketing,and has received extensive attention from the academic and commercial areas and has been deeply studied.However,current social network influence maximization is mostly make use of social network structure to measure the influence of a user to obtain the seed collection.While the difference in influence of the same user in different topics is not fully considered,resulting in the lack of association between the obtained seed nodes and the information to be transmitted.Based on the analysis and summary of existing works,this paper take weibo kinds of social network as the main research object,makes use of the characteristics of the structure and communication of information in social networks to study the problem of social network influence analysis on topics.The specific research results include:First,in view of the topic detection problem in social networks,this paper proposes a topic detection method based on global vector and LDA.For weibo social networks,the number of words is relatively small,topics are simple,and word items are sparse.This paper aims at the characteristics of short text of weibo content in short text.By training the similarity relation between words in the weibo corpus that is tagged,and then replacing the words with same tag and the similarity greater than the threshold,we get the text and the word list,and take them as the input of the clustering algorithm to get the text topic.The experimental results on real social network data sets show that this method can effectively reduce the latitude of the model text,reduce the confusion of topic detection,and improve the effect of topic detection.Second,aiming at the issue of influence propagation in social networks,this paper proposes an improved epidemic model combining the characteristics of linear threshold model and infectious disease model.In this model,the influence of an infected node to the vulnerable node is positively related to the intensity of the relationship between the two,and the influence of the infected nodes on the same easily infected node can be accumulated.When the node's activation threshold is reached,the node is activated,and the infected node is restored to the immune node with a certain probability.Experiments on scale-free networks show that the model can effectively model the influence propagation process.Third,in view of the problem of maximizing the influence of social networks in the topic area,this paper proposes a heuristic influence maximization algorithm.In the process of screening seed nodes,first of all the selected seed node sets are set up according to the ranking order,then the influence of the alternative nodes on the topic is analyzed,and the seed nodes are searched by the maximum coverage.Experimental results based on large scale data show that the seed set obtained by this method can activate more nodes and have a larger impact.