Based On Accelerated Proximal Gradient Method And Semantic Spam Text Message Classification

With the rapid development of Web2.0, more and more consumers shop online. Some users, forsome interests would make some spam comments, deliberately or denigrate certain products bytouting these fake reviews in reference to a certain extent, which affects the value of comments’information so as to confuse and mislead potential customers. How to effectively identify theinformation and remove fake reviews from the review texts to keep the real product reviews? Thisthesis identifies fake reviews from two respects of reviewers and comments.Firstly, we propose a spam detection method based on accelerated proximal gradient method.From the behavior of review spam purpose of starting user ratings deviant behavior patterns weredetected as spam indicators. This thesis taken from Amazon Site500records to a different reviewerscores for different commodities constitute a two-dimensional matrix, through accelerated proximalgradient method to identify deviation score, in order to find the most likely refuse reviewer.Compared with the traditional score-based detection methods can more accurately identify potentialreview spammers.General detection means to this end, they released the information to determine comment spam,but this has its limitations and are less accurate, according to the reviewer found not one hundredpercent score is rubbish commentators, therefore, this thesis has added a semantic text-based spamdetection system.Compared to the conventional detection system, only according to the similarityof the text to determine whether it is spam, this thesis reviews and merchandise based on the degreeof association and the individual words detection methods, but also consider the time ofbuying,comment time, user levels, user ratings and other factors. The experiment results show thatthe proposed method is better than the traditional text similarity-based detection method atidentifying fake reviews.