Overlapping Community Mining Combining Content And Link Data

In this work, we focus on the problem of overlapping community detection combining both content and link information. To mining the advanced complex structure of real network and surpassing tradition non-overlapping model, a lot of approach about overlapping community was proposed. These approaches consider a node in the network can simultaneously belong to multiple clusters. Most of these approaches only based one type of information-content or link, however, the characteristic of subgraph, which is an important feature in social network, is almost difficult to reflect from these approaches.To reflect the characteristic of community in reality, we perform our approach combining both content and link analysis and build our model in two steps. First, we transform the origin network into an Object-Attribute graph, establish a model of candidate subgraph based this Object-Attribute graph and define the relevance measure between the attribute/node and the candidate subgraph. Second, to find overlapping community, we propose an approach named Subgraph Overlapping Clustering for assigning the edges to corresponding suitable candidate subsgraph. At last, the two steps can bridge based a Bayesian-model in order to learn the underlying structure of network and find out the accurately overlapping clusters of community. Our experimental evaluation on two benchmark data sets and there result of experiment show that the proposed model significantly outperforms the state-of-the-art approaches for overlapping community detection.