Research And Implementation On Community Discovery From Network Based On Data Mining

As Web2.0 is booming, relationship between people and Internet has a great change in an increasing depth, especially the advent of the Social Network Sites, which has mapped the relationship and behavior in the real world to the Internet, brings us friends and human relations. However there still lies the problem that cannot meet user's social and information needs well in the current research and application, mainly shown as more emphasis on the visible and strong ties, less on the hidden and weak ones; more single relations, less multi-relations. After analysis of the disadvantages of traditional algorithms, in this thesis some improved methods were proposed for community discovery from hidden weak tie networks with the use of the data mining techniques, including community structure detection, association path mining and finding core members, at the same time took its applications into consideration.Owing to various communication ways, there appear multiple relations between any two users and therefore we studied the community structure detection and association path mining from multi-networks in this thesis. Firstly, community structure detection in multi-network could be solved by traditional methods in single one, but multiplicity of ties was still a problem. Thus an improved method was proposed for community structure discovery in extracted multi-network. As to association path mining, we had to distinguish it from the traditional methods in single ones for its emphasis on the relation types between two nodes in the paths and controllability of path length, and thus an algorithm based on Hidden Markov Model was proposed. According to the Six Degree Separation theory, our algorithm can work out the problems above. Finally, different from traditional core member discovery algorithm, one opinion leader was found in each opinion community with the measurement of Page Rank based on the network structure in the area of opinion mining.As is shown in our experiments, in co-interest multi-network communities detected are used to recommend persons of similar interest to target users, and association path is adopted to help acquire information; in opinion network opinion community discovered from network with a co-opinion weak relation, helps find all of the representative opinions on something, and opinion leaders found in each community are beneficial to understanding each representative opinion thoroughly.