Fake Product Reviews Identification Based On Deep Learning

With the rapid development of the e-commerce industry,the online shopping model is becoming more and more mature,and an online product review system has emerged.Customers can choose products based on product reviews,and merchants can also get consumer feedback in a time based on reviews.In the context of the era of big data,the number of product reviews has generally grown exponentially.As customers attach more and more importance to evaluations,some merchants have begun to speculate and hire brushers to give their products a lot of praise,which often misleads consumers select the goods that really suit them.Today,the quality of goods is uneven,how to filter out real and effective reviews from a large amount of product review data has become an urgent problem.However,due to the large amount of fake comments and high concealment,identifying them has become a difficult problem.Thanks to the widespread promotion of machine learning,we can use models to analyze and fit comments.However,the obtained model still has problems of low recognition efficiency and low recognition accuracy.Inspired by neural networks,we can use deep learning network models to solve these problems,this paper mainly works is the following three aspects:(1)In order to solve the problem of tedious and lengthy Chinese text preprocessing due to the unclear process,it takes a long time and the preprocessing data is not ideal,a text data preprocessing process framework is proposed,and the review data is preprocessed according to this process.In order to set a control group for the experimental performance of the deep learning model,a variety of feature extraction methods and classification models based on feature engineering were tested,and the best classification effect on the n-gram using the logistic regression model was obtained to 0.893.(2)In order to solve the problem that existing deep neural network models can only extract a single feature,this paper combines the advantages of convolutional neural networks and recurrent neural networks,and proposes a hybrid neural network recognition model based on parallel methods,and uses three different feature fusion method fuses the global features extracted using a recurrent neural network and the local features extracted using a convolutional neural network to obtain a text representation with both local and global features.The product review data is identified.Compared with CNN and Bi-LSTM,the hybrid model can obtain a higher recognition accuracy rate,reaching 0.903.(3)In order to solve the problem that the initial randomized word vector cannot fully express the corpus semantics,a large number of network corpora are used to train the word vector.The skip-gram model in Word2 Vec is used to train the pre-trained word vector,and the original deep model recognition accuracy generally improved,the best accuracy is 0.915..