GCMK: Detecting Spam Movie Review Based on Graph Convolutional Network Embedding Movie Background Knowledge

In recent years, the movie industry is booming and many consumers regard movie reviews as an important reference for choosing movies. In the mean time, more and more movie marketing teams glorify their movies and suppress rival movies of the same period by hiring spammers to publish massive misleading reviews. It results in the existence of a large number of spam movie reviews on the online movie review platforms, which greatly misleads consumers and seriously undermines the healthy development of the movie industry. At present, there is little research on spam movie reviews, whose method of detecting spam movie reviews mainly relies on the text and statistical features of reviews, ignoring the significance of movie background knowledge such as movie characters and plots. In this paper, we propose a novel method for detecting spam movie reviews, which uses a graph convolutional neural network embedding movie background knowledge (GCMK). Specifically, we firstly construct a directed heterogeneous knowledge graph by using the movie synopses and the high-quality long comments of movies. Then we use the graph convolutional neural network to obtain the embedded features of the movie background knowledge, use BERT (Bidirectional Encoder Representations from Transformers) model to extract the text features of reviews, and obtain the correlation vectors between these reviews and the corresponding movies by comparing the embedded features of the movie background knowledge and the text features of reviews. Finally, we fuse text features, user statistical features and correlation vectors to construct the detection model. The experimental results demonstrate our proposed GCMK method is more effective than the other state-of-the-art baselines, with an F1-score of 84.94%.