False Review Identification Based On Multi-feature Fusion

With the rapid development of Internet and e-commerce,more users have begun to rely on online reviews on e-commerce platforms to make purchase decisions.However,fake reviews have become a growing problem,misleading and confusing consumers.Therefore,how to detect and filter fake reviews has become a current research hotspot.Currently,most studies on fake review detection adopt traditional machine learning methods.However,this method is highly dependent on natural language processing and has weak recognition ability.Considering that fake reviews are often highly simulated,it is difficult to build an accurate recognition model only from single-dimensional review content features.The detection method based on multi-feature fusion can effectively solve the problem of inaccurate single feature recognition.The most common multi-feature detection method at present is to identify fake reviews by analyzing the content,sentiment,semantics and other aspects of the review.However,this method ignores the behavioral and temporal characteristics of reviewers,resulting in low accuracy.In order to improve the accuracy of fake review detection,this thesis studies how to use reviewer’s behavioral and temporal features to assist fake review detection.The main contents are as follows:First,this thesis combines text features,reviewer behavior features and time features,and proposes a multi-feature fusion false review detection model.This model uses the Bi LSTM network and Bert network,which are excellent in the deep learning model,to extract features of different dimensions for deep fusion during the model construction stage.In this way,the model can more comprehensively and accurately identify fake reviews.Secondly,this thesis relies on the Yelp Chicago restaurant public data set to conduct multi-stage comparative experiments.The method proposed effectively improves the ability to detect fake reviews.The experimental results show that compared with the existing research models,the method improves the classification accuracy,recall rate and F1 score by 5.4%,6.1% and 5.7% respectively.Finally,this thesis develops a fake review recognition system based on multi-feature fusion.In terms of detection ability,the advantage of this system is that it improves the recognition accuracy of false reviews based on the multi-feature fusion model in this thesis.In terms of user interaction,the system supports single and batch review detection when identifying fake reviews.Users can easily and quickly use the system to identify false reviews,so as to better protect their consumer rights.