Detecting Fake Review Based On Relationship

Online shopping reviews provide valuable information for customers to compare the quality ofproducts, store services and many other aspects of future purchases. However, spammers try tomislead consumers by writing fake or unfair reviews to confuse the consumers. Previous researcheshave used reviewers’ behaviors to detect spammers. But these studies are only able to identifycertain types of spammers. However, in reality, there are many kinds of spammers who canmanipulate their behaviors to act like normal one, and cannot be detected by these existing methods.Spammer can publish the reviews which similar to normal one easily, so it is more difficult todetect the true fake reviews. In this thesis, we obtain the credibility score of store by AHP, honestyscore of review by text and other features first. Then according to the relationship of reviewer,review and store, the result of classification are obtained by Logistic.In the review graph, we have three kinds of nodes: reviewer, review, store and identify theirinter relationships: a reviewer is more trustworthy if the person has written more honesty reviews; astore is more reliable if it has more honesty positive reviews from trustworthy reviewers; and areview is more honest if many other honest reviews support it. This is first time such detectingmethod and review graph is proposed in fake review detection field, and an effective computationmethod is developed based on the proposed review graph model, different from existing method,the computation method do not use an review text information. Experimental evaluation of themethod we proposed shows: it is able to detect more difficult and subtle spamming activities, and itis more effective than other method, especially the run time of calculation time.