ACSD: An Attributed Heterogeneous Information Network Based Spammer Detection Model for Community Question Answering

As a kind of more and more popular information-sharing platform, Community Question Answering (CQA) portals attract a number of netizens to participate and learn from each other on them. However, some users post misleading questions and answers to promote a product or service, which can distort the decisions of other users and degrade the credibility of the CQA environment. To address this issue, conventional solutions typically extract user features and classify them to detect CQA spammers. However, they often ignore the rich interaction relations among CQA objects. In this paper, to make up for this limitation, we propose ACSD, an Attributed Heterogeneous Information Network (AHIN) based Community question answering Spammer Detection model. Specifically, the Meta-Path based Neighbors (MPNs) defined in the AHIN are used to leverage the structural information among CQA objects, and meanwhile, a hierarchical attention mechanism is integrated in the ACSD model to assign the weights according to the different levels of importance for objects, MPNs, and meta-paths. We evaluate ACSD in a real-life CQA dataset, and the results demonstrate the effectiveness and advantage of this model on CQA spammer detection.