Research on fake reviews detection based on graph neural network

In this paper, we address the characteristics of various associations implied between fake reviewers and fake reviews, and the ability to construct review network graphs with dependency relationships. Firstly, we use graph neural networks to extract association features between reviews, propose a method to improve graph neural networks based on TrustRank and propose a multi-feature detection framework. Then the graph neural network is used to aggregate neighborhood feature output embeddings and connect them with the paragraph vectors extracted by Doc2vec. Finally, the experiments show that the algorithm in this paper can detect more fake reviews and has a better check-all effect.