基于代价敏感图卷积网络的虚假评论检测研究

Research on False Review Detection Based on Cost-Sensitive Graph Convolutional Network

虚假评论是指在商品、服务或其他产品的评论中,存在一些不真实或者夸大其词的评论.虚假评论的存在会对消费者的购买决策产生严重的影响.但现阶段虚假评论检测方法中,并不能很好的提取文本特征,同时也没有处理数据集中包含的不平衡数据.针对这一问题,提出了一种基于代价敏感图卷积网络的虚假评论检测模型.该模型考虑了评论之间的相似性和关联性,将每个评论看作一个节点,并通过评论之间的相似性构建了一个图.利用代价敏感函数来处理数据中的不平衡问题,调整分类器的损失函数以更好地处理不同类别的样本.实验结果表明,相较于传统的虚假评论检测方法,模型在精确率、召回率和F1值等指标上均有所提高.

Fake reviews are reviews of goods,services,or other products that are untrue or exaggerated.The presence of fake reviews can have a serious impact on a consumer s purchasing decision.However,at this stage,the fake comment detection method cannot extract the text features well,and the unbalanced data contained in the dataset is not processed.To solve this problem,a false comment detection model based on cost-sensitive graph convolutional network was proposed.The model considers the similarity and correlation between comments,treats each comment as a node,and constructs a graph from the similarity between comments.Use cost-sensitive functions to deal with imbalances in the data,and adjust the loss function of the classifier to better handle different classes of samples.Experimental results show that compared with the traditional fake review detection methods,the model has improved the precision,recall and F1 value.