Research On Social Network Community Detecting By Integrating With Topic Attributes

With the development of Internet, the scale of online social networks have become verylarge and complexity. The purpose of community detecting is to dig out the community structurefrom social networks. Structure-based community detection with the help of network topic ismore effective on analyzing community structure, preventing crime than the one without topicfeature. This research proposes a community detection framework based on topic analysis. Themain work of this paper consists of three parts.Firstly, we suggest the communication behavior based network constructing method tosolve the uncertainty problem of describing network topic rely on communication relationship.Then we construct each communication network along with each topic which comes fromextracting communication content. We demonstrate that this method that combine networkstructure and topic effectively provides a solid foundation on community detection.Secondly, we found a disadvantage about text content on the method of frequency-basedtext similarity measures and determine to use word similarity to substitute the previous.Traditional clustering methods cannot deal with the sensitivity of parameter and cannot beapplied to a large scale circumstance. To solve this problem, Hierarchical Affinity Propagtion hasbeen proposed. Experimental results have proved the availability of the proposed methods.Thirdly, from aspect of community detection, to solve the problem of being unable toguarantee efficiency and effectiveness of community detection based on local expansion, animproved algorithm based on seed community to assess fitness degree is proposed; to solve theobscuring problem raised from merging relation in multi-relation community detection, a methodto assess node relation strength of multi-relation network, and based on which to detectcommunity is also proposed. Experimental results demonstrate that our approach’s communityresults are more closed to the actual situations from network.Last but not least, the thesis designs and implements the proposed framework and applied itto the Enron email data collection, confirming the effectiveness and usability of the system.Thethesis also summarizes the study of the whole paper, thus building the determination directionfor the next step research.