Research And Implementation Of Sentiment Classification System For Chinese Commodity Reviews Based On Deeplearning4J

With the rapid development and maturity of Internet technology,Internet technology has greatly improved people's lifestyles.Now people can buy things from all over the world through e-commerce platforms without leaving their homes.While purchasing goods,they can also evaluate the purchased goods.Effectively processing and analyzing these review data is of great significance to guide merchants to improve products and help users make quick decisions.Therefore,more and more researchers have carried out research on sentiment classification of product reviews.The current mainstream sentiment classification research methods are mainly divided into rule-based methods and machine learning based methods.In the specific implementation process,the rule-based method cannot process texts that do not meet the specifications,and traditional machine learning methods need to define a large number of sentiment dictionaries and focus on the selection of artificial features,which has certain limitations.With the development of deep learning,more and more researchers choose to use deep learning related technologies to conduct sentiment classification research.Using deep learning can effectively improve the accuracy of classification and reduce labor costs.However,in the current product sentiment classification system,most of them simply display the classification results and the word frequency of the corpus,and cannot accurately extract and display the user's opinion words describing the product in the review data.There are many popular deep learning frameworks,such as TensorFlow,Torch,Caffe,Theano,etc.,these frameworks are basically developed based on Python or C /C++,while those enterprises that use a large number of open-source projects in the Java ecosystem for deployment need to solve cross platform problems when using these frameworks for deep learning related research and development.In view of the above shortcomings,the system uses the Deeplearning4 J open-source deep learning framework based on Java language for development,and the whole system does not need cross platform development.At the same time of emotional classification of Chinese commodity reviews,the words describing commodity features in commodity reviews are extracted and displayed in combination with the classification results.The main work of this thesis is as follows:1.Build a sentiment classification model of Chinese product reviews based on Deeplearning4 J open source framework.The model includes data collection module,data preprocessing module,comment emotion classification module,classification result evaluation module and viewpoint word extraction visualization module.From data acquisition and processing to visual display,users only need to input the URL address of the products to be analyzed in the online mall.2.Give the method of crawling Chinese comment data.This thesis designs a crawler program to complete the crawling of product comment data and product details,and uses Selenium ChromeDriver to operate the browser core to simulate the user to operate the browser to access the page.The relevant elements of the page are located by XPath to obtain the data to be crawled,and the browsing depth is constantly changed in the process of crawling,so as to crawl all comments of the product specified by the user.3.Give the emotional classification method of Chinese reviews.In this thesis,Deeplearning4 J open source framework is used for development.By adding the UIServer dependency,the current network state can be observed during the training process,and the training can be stopped in time to optimize the network.In this thesis,attention mechanism based bi-directional long-short-term memory network and convolutional neural network(AttBiLSTM-MCNN)are used to construct a classifier to complete the emotional classification of Chinese reviews.4.Give the visualization method for extracting opinion words.Through the comment opinion extraction interface of Baidu's AI open platform natural languageprocessing module,the opinion words appearing in the comment data are extracted and combined with the classification results for visual display.Different display methods are used for different types of data to make the results more intuitive,and users can learn related information faster.5.Implement and test the Chinese comment classification system.The whole system is developed based on the SpringBoot + MyBatis architecture,and other modules of Chinese comment emotion classification system are embedded.According to the overall process of the system and the functional structure of each module,the whole system is tested.In order to verify the effectiveness of the classification method given in this article,parameter selection experiments and comparative experiments were carried out on the crawled product review data.The experimental results show that when the number of convolutional layers is 3 and the convolutional windows are 3,4,and 5,respectively,the classification method presented in this paper has the highest accuracy rate,reaching 91.23%;at the same time,it is further compared with other classification methods Experiments,the experimental results show that the method presented in this paper is about 3% higher than the mainstream classification method,and the accuracy rate,recall rate and other evaluation indicators have been improved accordingly.Through a comprehensive test of the system implemented in this thesis,it is shown that the system can achieve a high accuracy and automatic emotional classification of Chinese reviews of goods,and can accurately extract the opinion words appearing in the review data.Through the visual interface of the system,users can intuitively understand the advantages and disadvantages of goods and sales volume,so that consumers can quickly make purchase decisions and merchants can improve the goods.