Acta Scientiarum Naturalium Universitatis Pekinensis ›› 2017, Vol. 53 ›› Issue (2): 262-272.DOI: 10.13209/j.0479-8023.2017.033

• Orginal Article • Previous Articles     Next Articles

An Individual-Group-Merchant Relation Model for Identifying Online Fake Reviews

Chuanming YU1, Bolin FENG1, Yuheng ZUO1, Baiyun CHEN1, Lu AN2,()   

  1. 1. School of information and safety engineering, Zhongnan University of Economics and Law, Wuhan 430073
    2. School of Information Management, Wuhan University, Wuhan 430072
  • Received:2016-07-22 Revised:2016-09-24 Online:2016-11-30 Published:2017-03-20
  • Contact: Lu AN

基于个人-群体-商户关系模型的虚假评论识别研究

余传明1, 冯博琳1, 左宇恒1, 陈百云1, 安璐2,()   

  1. 1. 中南财经政法大学信息与安全工程学院, 武汉430073
    2. 武汉大学信息管理学院, 武汉 430072
  • 通讯作者: 安璐
  • 基金资助:
    国家自然科学基金(71373286, 71603189)资助

Abstract:

A novel individual-group-merchant relation model is proposed to automatically identify fake reviews on E-commerce platforms, which focuses on the characteristics of fake reviewers’ behaviors instead of review contents. Three sets of indicators are proposed, i.e. individual indicators, group indicators and merchants’ indicators. To validate the model, an empirical study of fake review identification from a Chinese E-commerce platform is implemented. A number of 97804 reviews posted from 9558 different IP addresses, which are related to 93 online stores, are selected as test data. Results show that the F1-measure values of the proposed model on identifying fake reviewers, online merchants and groups with credit manipulation are 82.62%, 59.26% and 95.12%, respectively. Utilizing logistic regression and K nearest neighbor classifier based on the comments of the content as the baseline methods, the F1-measure values are 52.63% and 76.75%, respectively. Thus, the IGMRM model outperforms traditional methods in identifying fake reviewers.

Key words: credit manipulation, fake review identification, user behavior modeling, IGMRM

摘要:

从评论利益相关者内容与行为特征相结合的角度, 提出一种基于个人-群体-商户的主体关系模型(IGMRM)。选择93家店铺中9558个不同IP的97804条评论作为样本数据进行实验, 结果表明, IGMRM在识别虚假评论者、存在信用操纵的商铺以及虚假评论者群体的 F1 值分别达到 82.62%、59.26%和95.12%。使用基于评论内容的逻辑回归模型和 K 最邻近模型作为基线分类方法, 识别虚假评论者的 F1 值分别为52.63%和76.75%, 表明IGMRM在识别虚假评论者方面优于传统方法。

关键词: 信用操纵, 虚假评论识别, 行为建模, IGMRM

CLC Number: