Mobile Online Review Analysis Based On Deep Loop Network

The diversification of smartphone brands has led to fierce competition in the market,which has placed higher demands on the performance of the smartphone itself and all aspects of after-sales service.It is very important for each brand supplier to understand users’ concerns and needs,and to identify their own shortcomings and make efforts to improve them.Similarly,for consumers,only when they have sufficient knowledge of the strengths and weaknesses of each phone can they buy a phone they are satisfied with.Through a series of analyses such as sentiment analysis and subject word analysis of cell phone review data,we can fully understand the advantages and shortcomings of different cell phones and provide suggestions to cell phone suppliers and consumers.Based on the suggestions,each cell phone supplier can make subsequent updates and upgrades,and consumers can use them as a basis to choose a cell phone that better meets their needs.Therefore,this paper proposes a model based on deep recurrent network to classify and analyze the review data of cell phones,and dig out the advantages and shortcomings of each cell phone.For traditional convolutional neural networks,the size of the convolutional window is fixed when performing sentiment analysis on text data,and the setting of the window size is a challenging problem;too small a window will lose effective information,and too large a window will lead to too many model parameters.At the same time,convolutional neural networks can only focus on local features of text data,but not global features,which will affect the accuracy of classification.For the current recurrent neural network and its several variants,each word representation contains the information of all previous words and has a strong ability to capture the context,but such models have a bias that the later words are more important than the previous words,which is not consistent with common sense.Moreover,information can only be propagated forward in such models,resulting in the state of the current moment depending only on the information before that moment,independent of the information after that moment.To address the above problems of traditional models,this paper proposes a hybrid model that combines convolutional neural networks and bi-directional GRU networks-CNN-Bi GRU hybrid model-for text classification.The model combines the advantages of different networks and weakens the disadvantages of a single network.For the problem of setting the convolutional window that exists in the convolutional neural network,this model uses convolutional kernels with different window sizes for feature extraction in the convolutional neural network part respectively,which can enrich the extracted features and avoid the loss of effective information,but also avoid the appearance of too many parameters and reduce the computational effort.For the biased problem of recurrent neural network,this model uses bidirectional GRU instead of unidirectional RNN in the recurrent neural network part,which can effectively alleviate the biased problem of the model and reduce the computational effort.The features extracted by the convolutional neural network part and the recurrent neural network part of the hybrid model are fused to enrich the extracted features.By comparing the experiments with traditional machine learning models and neural network models on a known corpus,it is found that our proposed model has a significant improvement in performance.Finally,the review data of three best-selling smartphones in the first half of 2020,iPhone11,Huawei mate30 and Xiaomi 10 on Jingdong Mall,are obtained through crawlers,and a CNN-Bi GRU hybrid model is applied to sentiment analysis of the review data to obtain the positive and negative review data.The LDA topic model is then applied to determine the hot review terms in the review data,which in turn infers which specific aspects of the phone’s performance consumers are mainly concerned about when purchasing the phone.The word cloud graph analysis and semantic web analysis are also performed on the positive and negative review data respectively to obtain the advantages and shortcomings of different smartphone models,which provide reference for consumers and suggestions for manufacturers to improve their products.