Research On Fake Reviews Detection Technology For Commodities

With the rapid development of e-commerce,product reviews generated by online shopping provide important reference value for consumers to choose goods.However,driven by interests,some businesses will hire commentators to make fake statements about the attributes of commodities,so that consumers receive fake feedback from the outside world on the quality of commodities,thereby affecting consumers' consumption behavior.Fake reviews can influence business reputation and make consumers tendency of choosing goods.Therefore,illegal businesses have an urgent need for fake reviews,which leads to the increasing proliferation of fake reviews and undermines the good competition rules that the e-commerce market should follow.These phenomena above have made research of fake comment detection to become one of the most important tasks in the benign development of e-commerce.At present,researchers usually analyze the comment text and the commentator's behavior to detect the false comment,but few people analyze the abnormality of the comment basing on the business features,the commentator's historical behavior and the comment time.Therefore,the previous methods can only detect fake comments with low concealment,while the detection accuracy of fake comments with high concealment written by professional fake commentators is low.In order to solve these problems and provide more accurate reference for consumer shopping,this paper mainly carries out the following two aspects of work:(1)A fake review detection method based on habitual bias and XGBoost algorithm is proposed.Firstly,an improved algorithm for computing emotional polarity is proposed,and localized emotional words are added according to location factors to make the calculation of emotional polarity more accurate.Secondly,the concept of abnormal fluctuation interval of merchants and the user habit deviation index of reviews are proposed,and the feature model of review-reviewer-merchant is obtained by integrating multi-dimensional important features.Finally,combined with XGBoost algorithm,fake reviews are detected.Experiments show that this method can effectively detect fake reviews and provide more effective guidance information for consumers.(2)A fake review detection method based on topic model and commenter anomaly is proposed.According to the characteristics of reviews,the method divides fake reviews into two types: content-based fake reviews and deceptive fake reviews.Firstly,LDA topic modeling is applied to the experimental data set to detect content-based fake reviews with inconsistent topics;then,deceptive fake reviews are identified by reviewer anomaly.This method assigns a score to each comment according to different features and feature weights.Finally,the final score is obtained by adaptive weights based on similarity between the abnormal period and the reviewer.Reviews are considered false if the score is high,while reviews with low scores are authentic.This method avoids a large amount of computation.Experiments show that this method is an effective detection method.