A Study Of Edge Clustering Method Based On Affinity Propagation

There are many social networks presence in our lives, such as the relationship between people can be abstracted as social networks, the human brain neurons can be abstracted as social networks, every day people to participate in the discussion of Weibo can be abstracted as social networks. There is often a very complex structure in social networks, the study of these complex structures often makes sense. In recent years, with the research work of social network analysis of the deepening, social network analysis has become a hot issue in the field of data mining.For social networks, there are two main attributes, namely point and edge, the connection between them. Such as the network of relationships between people can be examined a node, the social relations between them can be examined the edge. In the dense social network structure is usually divided into two cases, one case is the node is not allowed to overlap and another case is the node is allowed to overlap. With the development of the study, researchers found that the case of overlapping nodes is often better able to explain the actual situation of social networks, that is, overlaps community discovery research, which is currently the main research directions.Overlapping community discovery was made by one of the research carried out by the node, there are also many research methods about the node. Later it was discovered, the edges of the community research often brings better results. With the emergence of LC method study, about edge community discovery has become an important community discovery research. After calculating the similarity edge, the LC clustering method based on hierarchical classification and segmentation method for density communities divided community, but it is easy to make the clustering results into local minimum, and using the density segmentation method will make the community is divided too small. For the above shortcomings, this paper uses Affinity Propagation clustering method to divide data, and combined with the greedy strategy and expansion modules Q_ov^E to improve LC method. Furthermore, proposed a study of edge clustering method based on affinity propagation. Experiments on several classical overlapping community social network data sets and synthetic data sets, then compared with the classical CPM method, i LCD method, while clustering LC methods, ALC method performes better in many evaluations.