Research On The Method Of Opinion Leader Mining In Microblog Community

Microblogging as a representative social networking of Web2.0 era,which has a great influence on the development of social networking services.Through the Microblogging service platform,the relationships between people become more closer.It has been a new choice of people's lives in cooperation and communication.In the process of gradual formation of microblogging network,the network structure also shows sparseness,and the internal members of a community structure are often highly correlated and connected,community detection can improve the quality of microblogging services.Label Propagation Algorithm(LPA)is a community detection algorithm with simple rules and low time complexity.However,the random selection of nodes has a certain influence on the stability of the final community structure recognition.At the same time,it is not applicable if users with multiple tags in overlapping community.So we proposes a User Core Label Propagation Algorithm(UCLPA)based on user core indicator,it can be used to discover multiple communities about topics.The ideas are mainly based on the Core-Periphery structure feature and multi-label propagation algorithm named COPRA(Community Overlap Propagation Algorithm).Firstly,through the indicator of the user core degree,we can get the core user set,then according to the propagation rule proposed in this paper that combines with the propagation coefficient to update the labels,and the topical communities are mined accurately and steadily.Opinion leader mining is an important research subject of social network,especially in public opinion mining,network marketing and other aspects of great significance.Currently opinion leader mining in microblogging usually just considers the user attributes,network structure or interactive features,but the characteristic of microblogging topic is considered less.However,opinion leaders tend to have domain restriction which they have greater influence on particularly certain topic areas but weak in other areas.To solve this problem,we propose TopicSimilarRank algorithm based on interactive information and similarity of topics,which can be called TSR algorithm in short.The algorithm consider s user attributions and text characteristic in Microblogging,building links between the users based on user interaction informations that combine with topic similarity to construct a directed-weighted network,then considering the idea of vote in PageRank algorithm to mine opinion leaders.In this paper,we select Sina Weibo data sets to our experiment,and the experimental analysis of the relevant overlapping community discovery algorithms are evaluated through the modularity index,and opinion leader mining algorithms are verified by the coverage and improved core radio index.The experimental results show that the UCLPA algorithm has a large degree of modularity,indicating that the quality of the community division is more accurate,moreover,time efficiency has been significantly improved.TopicSimilarRank and the related opinion leader mining algorithms are verified by the average coreradio on several theme communities,and the TSR results are higher,indicating that the TopicSimilarRank algorithm is more accurate for opinion leader mining in theme areas.