Research On Deceptive Spam Review Detection For Combining Multi-features

In the current information society,online review data can greatly influence whether people consume a product.Therefore,in order to gain greater commercial benefits and influence consumers' consumption choices,businesses may hire spammers to write deceptive spam review that promote their own products or depreciate competitors' products.In order to maintain a normal review environment,it is necessary to research how to detect spam review.At present,the main method is to detect spam review by designing and constructing feature classifiers,but the method does not extract review features from multiple angles and does not have a specific design detection model based on the characteristics of review features.This result in incomplete feature extraction or partial information loss in feature fusion,which cannot further improve the accuracy of spam review detection.Considering the problems in spam detection,this paper starts from two aspects: review text and review behavior.Based on review text,this paper analyzes the differences between non-spam review and spam review in text topic,text sentiment and text semantics,and then extracts the topic features,sentiment features and semantic features.Based on review behavior,this paper analyzes the behavior difference between non-spam review and spam review from three perspectives: behavior sender,behavior receiver and behavior carrier,and then extract review behavior feature.Based on feature extraction,this paper analyzes the hierarchical structure and characteristics of review features.By using the attention mechanism,this paper firstly fuses the topic feature,sentiment feature and semantic feature within the review text feature,and then fuses the review text feature and review behavior feature.The spam review detection model is designed by hierarchical fusion.This paper takes Yelp.com review data as an example to conduct experimental analysis on the detection model.Firstly,the influence of topic feature and sentiment feature of different parameters on the model detection effect is analyzed.Secondly,the detection effect of the model is compared and analyzed.Compared with the baseline model,the hierarchical fusion model proposed in the paper has a great improvement in experimental evaluation indicators.The experiment shows that the proposed model is significant and effective for the spam review detection.