Research Of Opinion Leader Discovery Technology In BBS

With the increasing complexity of the current social environment and the Internet network environment, it is more important to effectively detect and control the cyberspace public opinion. BBS as an important news and information dissemination channel in China has been an unneglect phenomenon, where the BBS opinion leaders play significant roles in formation and guidance of public opinion. Influence discovery research draws more and more attention from researchers.Based on carding the importance of Blog both at home and abroad, the importance of forum participants and the research results of opinion leaders discovery, this thesis proposes the measure indicators about opinion leaders in online forums and puts forward our research ideas and technology roadmap. This paper selects Tianya BBS as research subject. Concerning the shortage of traditional method for calculating influential topic, in this paper, we builds BBS opinion leaders recognition model in two aspects. First, from the perspective of the semantic, we proposes a concept named high weight words collection and on the basis of it, a new model is presented by calculating word's influence on re-comment chain, our model can discover the influential people in BBS community. Second, based on the assumption that the forum's re-comment chain reflects the structure of influence transmission, we investigate an online forum and construct a social network with the replying relations between comments mapped to the comment authors'relations. Inheriting the idea of PageRank, a new model which can discover the influential people in BBS community is presented based on the replying relations random-walk. At the same time, we propose that adopting in-degree information of each node as a new initial iteration vector can reduce the number of iterations. Experimental results show that the model generally can converge with about 50 iterations and the suppose that adopting in-degree information of each node as a new initial iteration vector can reduce the number of iterations is correct. This paper focuses on presenting the second modeling method.Compared the ranking results with the results obtained through social network analysis, correlation analysis proves that the average Pearson correlation coefficient of PR and inlink, PR and betweenness, PR and eigenvector are greater than 0.7 in both single-post experiment and multi-post forum experiment. They are highly linear positively correlated, which illustrates that the proposed modeling method and the social network analysis method are highly consistent. Through the above analysis, we proves the rationality and necessity in the design of the model. Finally we analysis the phenomenon of forum opinion leaders replacement and discuss the stability of the status of opinion leaders. It has laid a foundation for further study of the characters in the behavior of complex networks.