Spam Detection with Integrated Review Text and Reviewer Behavior

The user-centered nature of the Web 2.0 era has prompted some merchants to use spam reviews to manipulate ratings and attract users. Since the performance of traditional methods for identifying spam reviews needs to be improved, this paper reviews and summarizes the existing methods for identifying spam reviews, and proposes a spam review identification model based on the review text and reviewer behaviors. The model extracts feature from the user’s review text and behavior separately, generates multiple decision trees using the random forest algorithm, integrates features for classification, and finally realizes the identification of spam reviews. The experimental results show that although the text of the spam reviews gradually becomes more and more like real ones, the spammers usually have obvious behavioral patterns. Compared with the identification models using only textual features or behavior characteristics, this model has significantly improved accuracy. Meanwhile, this model does not require large training sets and has strong generalization ability.