Detecting Review Spammers Based On Review Feature

Online product reviews can significantly affect product sales, resulting in a large number of reviewers who promote and/or demote target products by writing untruthful product reviews. A heterogeneous review graph was proposed to capture the intricate relationships among reviews, reviewers and stores, which can determine the spamicity of each reviewer by using an iterative computing model.In this paper, we propose a new review graph structure, namely product review graph, to tackle the fake reviewer detection problem in a storeless shopping environment. Our proposed product review graph not only captures the relationships among reviews, reviewers and products, but also enriches the nodes in the graph with abundant features extracted from review data. We propose effective scoring criteria to measure the spamicity of each kind of nodes, which can lead to high detecting precision. An effcient algorithm(ICE) is designed to fasten the iteration process by eliminating a certain portion of reviewers during each iteration. The algorithm returns the result to analyze the distribution of the various features, come to different detection algorithms apply the type of spam reviewer.It is easy for the practical application of the algorithm.We further employ a pseudo supervised learning method as a postprocessing of ICE. This machine learning method not only verified the effectiveness of ICE, but also can identify more suspicious reviewers in addition to those identified by ICE. Experiments show that our methods can detect subtle spam reviewers, thus can be used as a complementary technique to existing methods.Developed a spam reviewer detection software, the software can quickly search for reviews and product information to facilitate.It is easy to evaluate the algorithm result and analysis the result of evaluation to ensure that the comments were spam reviewer detection study were real efficient.