Web 信息处理与应用WebMining6 SNAcommunity2.pdfVIP

Web/Networks Structure Mining: Communities One of the most important structural properties in networks 1 Community Detection Methods Connection between community detection and clustering  Agglomerative hierarchical clustering  Partitional clustering  K-means  Divisive hierarchical algorithm ––Divisive hierarchical algorithm Girvan and NewmanGirvan and Newman  Spectral graph cutSpectral graph cut  Modularity maximizationModularity maximization 2 Basic Graph Notation i Wij j  A graph G = (V,E) consists of vertices V and edges E  Edges can be weighted or unweighted.  The adjacency matrix邻接矩阵 A (weight matrix W) describes the graph:  A =0 (W =0) if vertices i and j are not connected ij ij  A =1 (W =weight of the edge), if they are connected ij ij  The degree of a vertex is the sum of all adjacent edge weights: d = SA (S W ) i j ij j ij  All vertices which can be reached from each other by a path form a connected component （连通分支） 3 Graph Partitioning  Undirected graph G=(V, E):  Partitioning task:  Divide vertices into two disjoint groups (A, B) A B 1 5 2 4 6 3  V=A U B  Questions:  How can we define a “good” partition of G?  How can we efficiently identify such a partition? 4 Graph Partitioning What makes a good What makes a good partition? clustering?  Maximize the number of within- 