Attributed community search considering community focusing and latent relationship

Attributed community search is to find a subgraph with some specific attributes online in terms of given vertices. It can help us retrieve information on a subgraph rather than the whole graph, thus enable down-stream graph search applications. However, it is difficult for users to specify exact query vertices if they are unfamiliar with the required graph. Most existing community search methods depend on the query vertices strictly and cause the searched community to shift from the truth community. Meanwhile, due to the incompleteness of original graph data, there exist many latent relationships between vertices, which may influence the search results. But most existing methods ignore these latent relationships and usually lead to a result with low F1 scores. Therefore, this research proposes an improved attributed community search method considering community focusing and latent relationships. We first build a structure attribute network embedding model to learn representations for vertices. Based on this model, the latent relationships are discovered and added to the original graph. Then a community shifting correction algorithm is presented to solve community focusing problem and achieve a more desired community. The experimental work on real-world networks confirms that our method can achieve better performance than existing methods.