Research On Identified Deceptive Comments Based On Convolution Neural Network

With the application and popularization of Internet technology, Online shopping gradually become a new form of shopping method and the comments number of online shopping is highlighted. Only commodity information released by the seller can't provide consumers a sufficient judgment because online shopping have the characteristic of information asymmetry. Which prompted most consumers reading the comments before making shopping decisions to reduce the risk of shopping. Because of this kind of behavior patterns, many businesses in their own interests driven issued a large number of deceptive comments to mislead potential customers and improve their sales volume. Which cause serious damage to the network shopping environment and order. Ordinary consumers cannot easily identify these deceptive comments because these have professional and deliberately. Therefore, we need to design an algorithm to help consumers identify deceptive comments.Convolutional neural network (CNN) inspired by the natural visual cognitive mechanism of biological, it is a classic neural network.Convolutional neural network can be directly through the training data to learn. It put the original data into a higher level and more abstract expression with some uncomplicated model. This paper aims at the problem that it's difficult to extract deceptive comments' characteristics.Analyzing and summarizing the existing classification algorithms. This paper put forward a text representation method: Subject2Vec and a dynamic eigenvalue filtering strategy. Based on this two innovations and convolutional neural network algorithm, this paper designs S-DCNN identification algorithm. S-DCNN improves the accuracy of identification of deceptive comments. In addition, this paper built an online product comments judgment system based on S-DCNN identification algorithm,JavaScript, HTML and other web technology.In this paper, we combine the application of specific scene puts forward new ideas for text categorization, and use the improved convolution neural network to solve uncertain and extract the effective features in traditional text categorization problem, improve the accuracy of the deceptive comments recognition, providing a new method to similar problems.