基于超网络的电子商务虚假评论者检测方法研究

Research on Fake Reviewer Detection Method in E-Commerce Based on Hypernetwork

摘要: 随着电子商务的蓬勃发展，在线评论已成为消费者购买决策和商家信誉积累的关键依据。然而，受利益驱使，虚假评论(水军)现象日益猖獗，严重干扰了市场秩序。现有检测方法多聚焦于评论文本内容或用户的单一行为特征，忽略了虚假评论者之间，尤其是有组织水军团伙的高阶群体协作模式。本文将科研合作网络中的超网络建模思想引入电商虚假评论检测，提出一种基于超网络的虚假评论者检测模型(HBCD, Hypernetwork-Based Fake Comment Detector)。该模型首先将“商品–用户”评论关系构建为超网络，其中每个商品及其评论者构成一条超边；随后，模型深度融合了评论的文本异常性、用户历史信誉度以及关键的超网络协作因子(即用户在多个商品上与同一可疑群体的共现强度)。通过在真实电商数据集上的实验表明，本方法能有效刻画水军团伙的隐蔽协作结构，显著提升对虚假评论者，特别是群体性虚假评论者的识别准确率与鲁棒性，为电子商务平台的信用治理提供了新的技术视角。

Abstract: With the rapid development of e-commerce, online reviews have become a key basis for consumers’ purchasing decisions and the accumulation of merchants’ reputation. However, driven by profit, the phenomenon of fake reviews (online water armies) is becoming increasingly rampant, severely disrupting market order. Existing detection methods mostly focus on the textual content of reviews or single behavioral features of users, neglecting the higher-order group collaboration patterns among fake reviewers, especially organized fraudulent groups. This paper introduces the hypernetwork modeling concept from scientific collaboration networks into e-commerce fake review detection and proposes a fake reviewer detection model based on hypernetwork (HBCD, Hypernetwork-Based Fake Comment Detector). The model first constructs the “product-user” review relationship as a hypernetwork, where each product and its reviewers form a hyperedge; subsequently, the model deeply integrates the textual anomaly of reviews, users’ historical credibility, and the crucial hypernetwork collaboration factor (i.e., the co-occurrence intensity of a user with the same suspicious group across multiple products). Experiments on real e-commerce datasets demonstrate that this method can effectively characterize the covert collaborative structure of fraudulent groups, significantly improving the accuracy and robustness of identifying fake reviewers, particularly group-based fraudulent reviewers, thereby providing a new technical perspective for credit governance on e-commerce platforms.