摘要
针对现有虚假评论检测方法未充分利用用户历史行为中蕴含的动态信息,首先利用时序分析模型从这些动态信息中挖掘能够刻画用户行为的动态特征;其次,融合这些动态特征与用户层面静态特征发现可疑用户,并将用户可疑概率传播至用户所发表评论得到评论可疑概率;最后,融合评论可疑概率与评论层面静态特征形成融合特征,使用PU-Learning分类策略实现虚假评论的检测。真实数据集上的实验表明,本文方法的性能优于现有方法。
Considering that the existing detection methods fail to use the dynamic information contained in user sequential behaviors for the identification of fake reviews, this paper suggests a novel fake reviews detection algorithm to integrate the dynamic information. Firstly, a timing analysis model is used to extract the dynamic features which could characterize the dynamic information; secondly, the dynamic features are integrated with user static features to find suspicious users, and then suspicious probabilities of users are propagated to the user reviews; finally, the reviews probabilities are integrated with review features to construct a feature set, and a PU-Learning strategy is proposed to detect fake reviews. Experimental results on real data sets show that the method in the paper is superior to the existing methods.
出处
《信息工程大学学报》
2016年第4期504-508,512,共6页
Journal of Information Engineering University
