Research On Sentiment Analysis For Electric Business Reviews Based On Deep Neural Network

With the rapid development of information technology and the increase of the number of e-commerce platforms,people use online shopping more and more frequently.The demand for expressing opinions and emotions on the e-commerce reviews is also becoming more and more intense,resulting in a huge amount of commodity comment text to be processed.Using natural language processing technology to analyze the text of e-commerce platform,and to excavate the emotional inclination contained in it,also become an important way for industry investigation and after-sales information feedback of manufacturers.Therefore,it is of great commercial value and social significance to study the text emotion analysis method for commodity comment.At present,the research on the analysis of text emotion in China mainly focuses on the micro-blog and film reviews,which has relatively little research on the sentiment analysis of e-commerce comments.In view of the above problems,this paper makes a work on the data source and use the web crawler technology to obtain the commodity comment of the classification of JingDong electric water heater.And on the basis of grasping the comments of corpora,by improving the existing sentiment analysis algorithm,is proposed based on deep learning model on e-commerce platform comments emotion classification,and the experiment result shows that model has better effect than the existing algorithm.The specific work is as follows:1.In view of the special page structure of e-commerce platform,this paper developed a network crawler project with multi-thread,agent pool and intelligent scheduling based on the Scrapy framework in Python.The project is able to capture the latest review data in accordance with the rules formulated,and prevent the anti-crawler mechanism and network abnormality of the e-commerce platform.The extracted data is persisted in the database after simple cleaning,as the data source of the experimental part of this paper.2.Research on the application of convolutional neural network in sentiment analysis.Study found that the e-commerce comments sentiment analysis task based on convolution neural network which have a good effect on short feature extraction but CNN wasn't given focus on the emotional tendencies of part of speech which gives a greater influence.This paper thus put forward in combination with characteristics of part of speech and word vector convolution pre-training neural network model(tCNN),the experimental results show that the model has a improvement on the performance of sentiment analysis tasks compared with basic convolution neural network structures.3.This paper studies the application of existing recurrent neural network in sentiment analysis task.A bidirectional gated recurrent unit AttBiGRU with attention mechanism was proposed,and then the tCNN and AttBiGRU model were modified and fused.The experimental results show that the combined tCNN-AttBiGRU model has higher performance in the essay classification task.