基于BERT-BiLSTM-GAT的人工智能生成电商虚假评论识别研究  

Identification of E-Commerce Fake Reviews Generated by Artificial Intelligence Based on BERT-BiLSTM-GAT

作　　者：李世勇[1] 杨铮铮 LI Shiyong;YANG Zhengzheng(Yanshan University,Qinhuangdao,Hebei,China)

机构地区：[1]燕山大学经济管理学院,河北省秦皇岛市066000

出　　处：《管理学报》2025年第3期557-567,共11页Chinese Journal of Management

基　　金：国家自然科学基金资助项目(71971188);教育部人文社会科学研究资助项目(22YJCZH086)。

摘　　要：为了有效识别与治理电商平台中由人工智能生成内容导致的虚假在线评论,构建基于主动半监督框架下的BERT-BiLSTM-GAT混合模型,提出一种有效的AI评论识别方法,并在多种数据集中进行实证分析,验证模型在应对多类别的商品评论、多种AIGC模型以及不同提示词时的灵活识别能力。研究表明:该识别方法在羽绒服数据集以及跨领域数据集中达到较好的识别效果,优于目前常见的深度学习识别方法,验证了模型的性能优势;同时,在AI评论与人工评论的特征存在较高一致性的情况下,仍具备良好的泛化能力与适应性。In order to effectively identify and manage fake online reviews caused by artificial intelligence generated content(AIGC)in e-commerce platforms,we construct a hybrid BERT-BiLSTM-GAT model based on an active semi-supervised framework,propose an effective AI review recognition method,and empirically analyze the model in a variety of datasets and verify the model’s flexible recognition ability when dealing with multiple categories of product reviews,multiple AIGC models,and different prompt words.The results indicate that the method in this study achieves better recognition results in the down jacket dataset and cross-domain dataset,which is better than the current deep learning recognition methods,and verifies the performance advantage of the model.At the same time,in the case of high consistency between the features of the AI review and the manual review,it still possesses good generalization ability and adaptability.

关 键 词：AIGC 生成式人工智能 电商平台 深度学习 AI信息污染 虚假评论识别 

分 类 号：C93[经济管理—管理学]