基于类不平衡和特征选择的两阶段垃圾评论检测方法

Two-Stage Spam Review Detection Method Based on Class Imbalance and Feature Selection

用户在电商平台购买商品的时候,其他用户对相关商品的评论起着重要引导作用。出于影响用户购买倾向等目的,部分商家在电商平台存在恶意刷评论的行为。已有的垃圾评论识别研究重点从用户的购买行为等方面进行数据挖掘,目前还没有研究人员从中文电商平台的垃圾评论内容角度展开研究。从国内某一知名电商平台抓取相关数据,根据行为模式确定强疑似垃圾评论;针对搜集的数据集内存在的类不平衡问题和维度灾难问题,设计出了一种两阶段垃圾评论检测方法。实证研究表明,该方法构建的模型相对于仅考虑类不平衡或仅考虑维数灾难的基准方法,具有更好的分类效果。

Relevant commodity reviews play an important role in users’purchase through e-commerce platforms;therefore,some merchants have maliciously spammed reviews on e-commerce platforms for the purpose of influencing the users’purchase predisposition.The existing spam recognition research focuses on data mining from users’purchase behaviors,leaving the room for the research of spam reviews on Chinese e-commerce platforms.This paper captures relevant data from a well-known domestic e-commerce platform to determine the strong suspected spam reviews based on users’behavior mode,and then develops a two-stage spam detection method in view of the class imbalance and the dimension disaster within data sets collected.The empirical study indicates that the model constructed by means of the proposed method shows a better effect in spam review detection compared with the fiducial approach considering only the class imbalance or the dimension disaster.