A Fake Reviews Detection Model Based On Convolutional Neural Network And GRU

With the rapid development of the internet and e-commerce platforms,product reviews have become an essential part of e-commerce platforms,which greatly influences consumers’ purchasing decisions and are also used by businesses to evaluate brand perception and customer satisfaction levels.However,as time goes by,fake reviews have gradually increased on e-commerce platforms,misleading consumers’ purchasing decisions and damaging to the rights and interests of consumers and legitimate businesses.Therefore,detecting and filtering fake reviews is of great practical significance.In this paper,based on convolutional neural networks and GRU,the CNN-GRU model was proposed,which combined the advantages of both algorithms.Then,the Glo VGG-GRU model was proposed,which combined transfer learning and the VGG＿10 model that using an improved model VGG on the basis of CNN.The experimental results show that the proposed model had better detection effect on fake reviews.The specific work is as follows:(1)A method for fake review detection based on deep learning was conducted for the limitations of machine learning methods,such as poor handling of high-dimensional feature processing.Firstly,the characteristics of fake reviews on real websites were analyzed,and the dataset were preprocessed.Secondly,the input data using Doc2 vec to extract the word vectors of the text,so that retain the order information and preserve contextual information.Then,the CNN-GRU model was constructed,which used CNN for text feature representation and combined GRU for comment text classification.Through experiments comparing the performance of different deep learning models in fake review detection tasks,the model structure most suitable for fake review detection tasks was obtained.Finally,the effectiveness of the CNN-GRU model was verified by performance comparison with common machine learning algorithms and basic neural network models.(2)A disadvantage in the CNN-GRU model is that the number of layers not large enough and the extracted features are simple,when CNN processes fake reviews text.To solve this problem,the CNN-GRU model was improved using VGG networks.However,the performance of the classical VGG network is limited when using too small a dataset because of problems such as the network layers being too deep and the large number of parameters.Thus,a simplified convolutional neural network VGG＿10 structure was proposed for fake reviews detection,which was an improvement to the existing VGG16 model by analyzing the VGG network structural.(3)The Glo Ve model was used in transfer learning to solve the problem of over-fitting due to the small number of fake reviews dataset,which improved the identification of fake reviews.Then,the Glo VGG-GRU model was proposed,which combined transfer learning,VGG＿10 and the GRU model.Experiments were designed to analyze and compare the fake reviews detection results,and the results show that the accuracy,recall and F1 of the proposed Glo VGG-GRU model are 89.76%,87.69% and 88.71%,respectively,which was an average improvement of 13.93%,11.34% and 12.64%in these three metrics relative to the VGG model,GRU model,CNN+GRU model and VGG+GRU model,respectively.In summary,the Glo VGG-GRU model,which combined the improved VGG network,GRU model and transfer learning,performed well in comment text classification on real datasets and could provide a new method for detecting fake reviews.