Micro-Blog Short Text Sentiment Orientation Analysis Based On Deep Learning

With the rapid development and progress of social network media,the text information on the Internet has expanded dramatically.As one of most popular microblog services in China,Sina weibo generating huge amounts of data all the time.These data contain a lot of information about the emotions and attitudes of netizens.We can acquire huge value from the massive data if we apply sentiment analysis technology to them.The application of sentiment analysis on Weibo's massive data will help improve the Internet's public opinion monitoring system to detect abnormal or unexpected events in the physical world.This paper uses Chinese short text microblog as the main research target.The main research work of this paper includes:(1)In the aspect of microblog data acquisition.In this paper,we designed a crawler architecture,and based on this architecture and implemented a microblog data acquisition system written in python language.The system uses selenium to simulate the interaction of Weibo,and obtains the Weibo cookie to solve the identity authentication problem by simulating the human operate browser.At the same time,reference to the design pattern of the focused crawler,targeted capture data of Weibo.The crawler system solves the problem of complicated implementation of reversing microblog,at the same time,it solves the problem of too slow in applying completely simulation by using selenium,and can acquiring data efficiently.Ideally,using Single thread the crawler system can collect about 800 lines per minute.(2)In the aspect of sentiment analysis,we considering the positive influence of emotional words on the emotional expression of the text.In this paper,we integrated several Chinese sentiment dictionaries commonly used in Chinese sentiment analysis,and designs a strategy for weight adjustment of the word vector using emotional words.In order to verify the effectiveness of the method,we separated the unadjusted word vector and the adjusted word vector into TEXTCNN and LSTM respectively,and compares it through multiple experiments.The result shows that after adjusting the word vector TEXTCNN can achieve the highest classification accuracy of 84.1% in 2-way sentiment analysis task which is 2.1% higher than that without the word vector adjustment.It proves that the adjustment of the word vector has a positive impact on microblog sentiment classification task.(3)At last,we integrated the above research results,designed and implemented the background system of microblog emotion classification.The system provides an interface for collecting personal information of Weibo users,an interface for collecting keyword information,and provides an interface for emotional analysis of Weibo,which can achieve efficient acquisition of Weibo data and simple analysis of Weibo emotions,which laid a good foundation for subsequent research.