Abstract
Aiming at the limitations of existing spammer group detection approaches, we propose a novel approach for detecting spammer groups based on hypergraph embedding and autoencoder-classifier. First, we use time periodicity and rating behavior to model the review dataset as a heterogeneous hypergraph. Second, we combine the autoencoder and attention mechanism to design a hypergraph neural network to learn the high-quality low-dimensional vector representations of nodes from the heterogeneous hypergraph and use the MeanShift clustering algorithm to obtain candidate spamming groups by filtering items, ratings, and time nodes. Finally, we apply an abnormal detection model based on an autoencoder-classifier to detect the spammer groups. The experimental results on real-world review datasets show that the proposed method has better detection performance than the baselines.

This publication has 22 references indexed in Scilit: