Chinese Sentiment Analysis Based On Deep Learning

With the rapid development of Web2.0,the social networks have become an indispensable part of people's lives.People began to express their opinions on the network,generating a large amount of text data with emotional information.The use of sentiment analysis technology to mine text and find the emotional information in the text is of great significance to manty industries.Therefore,sentiment analysis has become one of the most active research fields in Natural Language Processing.In recent years,deep learning has been widely used in the field of Natural Language Processing,and has achieved good results.This thesis studies the Chinese text sentiment analysis method based on deep learning.The main work includes the following aspects:1.The web crawler technology is studied.The web crawler scheme is designed according to the characteristics of the website.Parallel crawling of data through the Scrapy framework.The crawl data is formatted and stored in the MySQL.The large-scale Chinese text sentiment analysis corpus is constructed.2.The generation of emotional word vector is investigated,and an effective method for generating emotional word vection is proposed.The method mainly adds emotion information to the pre-training word vector to obtain the emotion word vector.The specific implementation method is based on the constructed lange-scale Chinese text sentiement corpus training word vector,and generates the sentiment dictionary through the corpus and SO-PMI method,and then obtains the sentiment word vector by splicing the word vector with the emotional score of the word.In this thesis,the method is validated on three pre-training word vecotrs.The experimental results show that the application of this method can improve the effect of sentiment analysis task.3.A new network architecture,BLSTM-MultiAtt-CNN,is proposed.It includes BLSTM layer,multi-channel self-attention coding layer and attention CNN layer.Firstly,the sequences of statements are encoded by BLSTM layers,and the context information is captured.Then,the global information is obtained by the multi-channel self-attention coding layer,and the local information is captured by the attention CNN layer captures.Finally,the global information and the local information are spiliced together as a vector representation of the emo.In this thesis,the optimal values of hyperparameters in the model are determined by multiple sets of parameter comparison experiments,and the effectiveness of the model is verified by a comparison with various models.4.A RESTful API sentiment analysis service based on BLSTM-MultiAtt-CNN model is implemented.This thesis uses the Flask framework to build a Web system.It integrates the model,provides RESTful API services,and displays the service through visualization.Finally,it packages the service through Docker for easy deployment.