Rankclus On Directed Graph And Its Application

Networkstructuresemergealmostubiquitouslyincomplexsystem.Bothrankingandclustering can provide overall views on information network data,and each has been a hottopic by itself.However,applying either of them over information networks separately of-ten leads to incomplete,or sometimes biased results.A novel clustering framework calledRankClus is proposed to integrate clustering with ranking and the quality of clusteringand ranking can be mutually enhanced.Many work about mining structure of undirected network has been proposed, butthere are very few studies about the directed network.The asymmetry of the adja-cency matrix in directed graph causes essential difficulties for many clustering algo-rithms.However,RankClus algorithm can be easily extend to directed graph and becomea feasible method for mining structure on directed graph.Journal citation network is a real directed graph with both community struc-tur(eacademic sector) and ranking structure(impact).Applying RankClus algorithm,wecan mine the real structure of the network,including the time evolution.On the otherhand,discussion can be made about the effectiveness,convergence of the algorithm.The main innovation and contribution of this thesis are summarized as below:Extend RankClus algorithm to directed graph,clusteing and ranking on directedgraph；Apply the model and algorithm over journal citation network,mining the real net-work structure；Make some discussion about the algorithm according to computation；...