E-commerce Online Fake Product Review Detection System

In recent years,fake product review appearing quickly.These fake product reviews or bad reviews have seriously affected the fairness of e-commerce websites.Since fake reviews are hidden in a large number of real reviews,the existing research methods only process fake product reviews as an isolated static problem,and it is difficult to efficiently and effectively discovering fake reviews.For resolve above problems,the system mines the relationship among users and finds out suspect reviewers by analyzing the reviewer's behaviors and review's semantic,and evaluate suspect reviewers based on the corresponding semantic evaluation model.In order to be able to accurately discover fake review,this paper has done the following work:1.Research the basic business knowledge of e-commerce websites and source of fake product review.Study the current popular methods and solutions for solving such problems,as well as the usage scenarios,advantages and disadvantages of these methods and solutions.Based on detailed investigation and research,requirements and system goals were consolidated after elaborate investigation.Data processing subsystem,suspect user detection subsystem and suspect user management subsystem were designed to achieve the goals of this paper.2.In order to process big data and resolve data decentralized and data incomplete problems,Establish a data processing subsystem for collecting and processing review related data.The data processing subsystem is mainly used to integrate data from various external systems.Collection of reviewer profiles data,the processing of reviewer behavior data and the statistics of user transaction data were completed by triggering multiple Spark jobs in parallel.3.In order to resolve fake review detect difficultly problem and the anti-reconnaissance ability improving problem,the paper analyzes the motives of adding fake review,and establishes the connection between motivation and reviews.Innovatively use mixed review coverage algorithm,behavioral time difference method,and edit distance are used for mining suspect reviewers.And,fine-grained product review sentiment analysis is used for analyzing user emotion and intention,the calculation result would be the possibility of fake reviews.Finally,establish a credit evaluation model,apply user's behavior data and review semantic analysis results to the model,and comprehensively evaluate the suspect users.4.In order to resolve the problem of scattered and decentralized evidence of fake review,the suspect user management subsystem was developed to display the basic information,transaction information,evaluation results and evidence of the suspect users for business operator to verify the suspect users and their fake review,and finally determined the suspect's reviews as fake reviews.Suspect user management subsystem was built with Spring and ScrollSpy to present relevant data from multiple dimensions,enabling business operators to efficiently find the evidence that is needed to determine fake review and reviewers.Through testing and running on production,the results show that the system can collect the required data accurately and efficiently,and has the ability to discover most of the suspected users.The business operator can easily determine the suspects and their fake reviews through this system.This system greatly cleaned the e-commerce website's transaction environment and improved the fairness of the transaction process.