Research On Review Spam Detectiocn Algorithm

Review Spam on e-commerce platform has had a serious impact on the experience and vital interests of the customers. As a result, how to detect the review spam has become a hot research direction. The review spam can be divided into content-oriented and mendacious spam. The role of content-oriented spam comments is message propagation and the role of mendacious spam comments is changing the impression of the potential customers on the goods. This thesis analyzes and detects the content-oriented and mendacious spam by analyzing the forms of the review spam.The content-oriented spam often adds advertisement links, web spam, malicious communication information and pornography information. The function of this type of spam comments is to push information and spread virus. This thesis, which researches the forms and analyzes the feature of content-oriented spam comments, utilizes the BP neural network model and the combined model of autoencoding and nave Bayes model to detect the content-oriented spam comments.In the mendacious spam, the publishers of the comments utilize mendacious shipping information and order form to obtain the permission to comment and puff and defame the impression of certain object deliberately. This can be used to change the buying behavior of the potential customers. The mendacious spam comments influence the buying behavior and damage the right to know of the customers. This thesis proposes the detection model of mendacious spam comments based on the multi-characters. The behavior features of the publishers are very important to the detection of mendacious spam comments according to the research. As a result, this thesis uses the stacked autoencoders to learnning the behavior characteristics of the publishers and aims to learnning the behavior characteristics of the publishers more fully.Then we use the multi-characters to detect review spam.