Incorporating Attribute Information into Joint Random Walk for Interpretable Local Community Detection
Tianchao Chen, Huifang Ma, Ju Li · 2024
Local community detection on the attribute graph aims to locate the community that contains query nodes and satisfies structure cohesion and attribute homogeneity. Most of the traditional local community detection methods focus on topology information and fail to capture attribute information, making it difficult to capture dense subgraphs with high similarity of attributes. Aiming at the above issue, an interpretable Local Community Detection method based on incorporating Attribute information into joint Random walk (ARLCD) is proposed. Specifically, a heterogeneous graph is constructed to capture multiple of structural information. Then we leverage the spectral properties of nearly uncoupled Markov Chain to prolong the influence of user's past preferences on the successive steps of the walk, allowing the walker to explore the underlying features more comprehensively. In addition, to eliminate the seed dependence of random walk, we design a joint random walk with reinforcement mechanism. The experimental results on several datasets show our proposed method can find the community with high structure cohesiveness and attribute similarity.