Topic-aware Most Influential Community Search in Social Networks

Community search is a problem aimed at searching for densely connected subgraphs within a network based on query conditions, which has recently attracted significant attention. However, most previous community search studies have overlooked the coexistence relationship among attributes. They typically assign a single attribute to each node or edge (e.g.,only considering influence scores or keywords), which is difficult for users to obtain a comprehensive and beneficial information. Additionally, most of them also ignored the uncertainty in the attribute graph. Therefore, in this paper, we introduce two novel community models, namely topic-based interaction graph and $(k,l,\eta)$-influential community. The former is a directed ucertain graph generated by the query topic distribution provided by users, while the latter is used for solving the topic-aware most influential community search problem in social networks. Furthermore, we propose an online search algorithm which computes the influence value of each vertex by considering the topic-aware information diffusion process on interaction graphs. And then, we use a peeling-pruning strategy to iteratively find the topic-aware most $(k,l,\eta)$-influential community. To further speed up the search performance, we devise two lightweight index structures which efficiently support the search for the topic-aware most influential community within an optimal time. We also propose three optimization methods to improve the space and time costs of the index-based approach.