|
Abstract:
|
The analysis of social networks has drawn much attention, especially on the community detection problem. Many observed networks contain both the network information and the node covariates information, but current methods mainly focus on the network part only. To solve the problem, we proposed a new community detection method that combines both the connections and the covariates. On the product of the adjacency matrix and the attribute matrix, we apply the attributed-SCORE algorithm, to get reasonable community detection results. Our numerical analysis show that with our method, the covariates improve the clustering results even when they do not share the same community structure with the connection matrix. It is supported by our theoretical analysis. We apply it to the statistician citation network with the abstract of each paper as the attribute, and compare with the clustering results without the abstracts.
|
