Research On Fake Reviews Detection Based On Multi-dimensional Features And Deep Learning

With the rapid development of the electronic commerce,consumers are increasingly beginning to go shopping online and present their reviews regarding to products.In order to make right decisions,consumers get information not only from merchants but also read reviews about products.However,driven by profit,merchants employ multiple spammers to produce positive reviews.Those reviews attract more potential customers to buy their products.What's more,They hire spammers produce negative reviews of competitor's products so as to denigrate their competitors.The spammers' activities and fake reviews mislead potential consumers and are not conductive to the stable development of e-commerce.Therefore,we need to explore effective methods to detecting deceptive fake reviews.Based on the content of reviews and behavior of reviewers,this paper extracts multi-dimensional features and uses machine learning techniques to identify the fake reviews.Furthermore,this paper propose a novel model named as DF-HAN.The model extracts deep semantic features with deep learning techniques.Those semantic features are combined with multi-dimensional features for detecting fake reviews.The main contributions of this paper are concluded as follows:(1)Five feature indicators are proposed.Based on the content of the reviews,four kinds of text features are extracted,including N-gram features,words features,readable features,topic features.By considering the behavior of the reviewers,behavioral features are provided.Finally,five indicators are provided by analyzing the behavioral features and four kinds of text features.(2)Explore a model based on traditional methods for detecting fake reviews.In this work,five feature indicators are combined and LR,SVM,RF are used for developing the model.Experimental results show that The F1 value is 87.37%,which verifies the significant of the model proposed in this paper.(3)Develop a model of DF-HAN based on deep learning methods for identifying fake reviews.The model learns deep semantic features with LSTM and bidirectional GRU.The model also learns traditional discrete features,which are multi-dimensional and based on CNN.Then,semantic features and traditional discrete features are combined using DF-HAN model.Finally,the model of DF-HAN gives a higher accuracy(91.76%)compared with the model of HAN(85.99%),which shows some relative effectiveness of the model proposed in this paper.