Detection Of False Comments Based On Convolutional Neural Network

With the rapid development and application of network and electronic terminal technology,e-commerce has made rapid progress,once subverting people’s shopping methods.Online shopping has become a common way of shopping.When deciding whether to buy goods,many consumers not only check the details of the merchants’ introduction,but also often check the user’s product reviews and shopping experience,which brings about the problem of false reviews.Some bad businessmen employ a large number of false commentators to blindly praise the commodities and maliciously slander the commodities of their counterparts in order to seek unfair interests.This practice has seriously affected the shopping order of e-commerce platform,destroyed the fair shopping environment,and affected consumers’ shopping choice and experience.False reviews are deliberately written on the basis of merchandise information provided by merchants.They are professional and confusing and difficult for consumers to distinguish.Therefore,in order to maintain the stable development of e-commerce platform,it is urgent to study a reliable method to effectively identify false reviews to help consumers identify false reviews.Convolutional Neural Network(CNN)is a classical neural network inspired by the biological natural cognitive mechanism.It can map the comment data to a higher dimension for abstract expression,and further study the internal relations and rules of the comment data,so as to detect whether the user comment is true or not.In view of the lack of false comment corpus,based on Ott gold data set,this paper extends the gold data set by crawling some users’ comments on Amazon shopping website through the crawler tool.In order to improve the performance of false comment detection,a convolutional neural network model based on embedding mechanism is proposed.Specific processes include:(1)training the original data to the input layer of the convolution neural network by word 2vec;(2)using the convolution layer and the lower layer(pooling layer)to extract features and reduce dimensions,and flatten layer to flatten layer to flatten layer;(3)putting the original data into the full connection layer to distinguish the results.In this paper,experimental methods are used to study the structure training data of convolutional neural networks at different levels,and to find the best model for the realization of convolutional neural networks at the best level.In view of the over-fitting problem in the model,the corresponding parameters are adjusted in time to make the model work best.At the same time,the original data are trained into word vectors of different dimensions and fed into a comparative experiment of the same convolution neural network model.The influence of word vectors of different dimensions on convolution neural network is discussed.Finally,using the best training model,the extended golden data set is used to verify the performance.The experimental results show that the recognition performance of convolutional neural network for false comments is improved by dimension optimization,hierarchical structure optimization and data set expansion.