Research On Review Spam Detection Method Based On CNN Network Optimization

At present,with the rapid development of Internet information technology,online shopping has gradually replaced physical store shopping as one of the main shopping methods in daily life.When users buy goods online,the content of product reviews becomes the main reference information for users when purchasing goods one.In order to increase the sales volume of their own products or suppress the sales volume of the products in the competitor's stores,some merchants publish review spam of products to deceive consumers,which seriously affects the credibility of e-commerce platforms.Due to the large amount of product review information data,relying on manual identification of these review spam is time-consuming and labor-intensive,and the use of traditional machine learning methods can only mine the shallower and basic explicit semantic features of the review content.In response to this problem,this paper uses deep learning methods to dig out the deep hidden semantic features of the comment content and detect review spam.First,to solve the problem that traditional machine learning methods cannot extract the deep semantic features of review content,this paper uses the convolutional neural network in the deep learning model to mine the deep semantic features of review content,and the classification results are obtained through the output layer to detect review spam.This method constructs a certain level of convolutional layer and pooling layer in the convolutional neural network to form the main path convolutional neural network.On this basis,the structure of the convolutional neural network is optimized and improved,and a new side road model is formed by extracting each layer of the main road convolutional neural network to form a new model.Reinforce the subsequent learning and update the side road weights to strengthen some target samples that are classified incorrectly due to unobvious features,adjust the parameters,and detect review spam through the output layer classification.Secondly,in order to solve the two problems that the convolution kernel of the convolutional neural network in the aforementioned detection method is determined by the random initialization of the system and the labelled review data is insufficient,this paper uses a sparse autoencoder to pre-train the convolutional neural network in layers The method to solve these two problems,and then detect false comments.This method utilizes the feature that the autoencoder can restore the output data to the input data,extracts the preliminary features of the comment content through its hidden layer,and divides the extracted features into hierarchical pre-trained convolution kernels of the convolutional neural network.The convolutional neural network pre-trained by the autoencoder can extract the deep invisible features of the comment content to dig out more valuable semantic information,and at the same time can accelerate the convergence of the network and improve the accuracy of detecting review spam.Finally,experiments were conducted on the Amazon?cn dataset and Yelp dataset,and compared with existing methods to verify the effectiveness of the proposed method.