Sentiment Analysis Of Weibo Based On Deep Learning

The development and maturity of the information age has promoted the rapid development of social media.More and more network users are accustomed to freely express their views,opinions and attitudes on the internet through micro-blog.The massive information data accumulated on micro-blog platform contains huge emotional information and obvious public opinion tendency of users.In the context of the big data era,it is of great significance to accurately and quickly discover emotional tendencies through the deep learning technology for the formulation of enterprise marketing strategies and the monitoring of public opinion.Deep learning is a supervised learning algorithm,which constructs a multi-level nonlinear network model through constant parameter adjustment.Deep learning has a strong ability to mine potential information of data.Through the analysis of algorithm theory and the research on practical application,it is a feasible direction to analyze the emotional tendency of micro-blog based on deep learning theory.This paper introduces the basic principles and algorithm flow of microblog emotion recognition and the construction of micro-blog emotion analysis model based on deep learning.The main work includes:(1)The open evaluation data of micro-blog sentiment analysis provided by COAE2014 and the micro-blog commentary data collected by automatic crawler technology are used as the original data of later model learning.Through experimental comparison,it is found that in dealing with the imbalance of types and numbers,the effect of deep learning algorithm is better than that of shallow model.(2)We perform data pre-processing and word segmentation for micro-blog commentary sentences,and transform the emotion recognition problem of sentences into the problem of emotion recognition of lexical sequences,In order to improve the accuracy of lexical affective discrimination and reduce the intensity of experimental work,this paper uses the data visualization method based on "word cloud" to screen and judge specific vocabulary.We use the Word2 vec to extract the feature information of the original text data and construct the word vector matrix,so as to the training data of the subsequent sentiment classification model was obtained.(3)By using shallow learning algorithm support vector machine and random forest for modeling,and the deep learning algorithm LSTM is used to train the data,we get the classification model of emotional polarity of microblog.By comparing the performance indicators of the three models,it can be found that the deep learning algorithm can effectively retain the sentiment information in the microblog commentary,and the accuracy is improved by more than 5% compared with the shallow learning algorithm,which has obvious advantages over the shallow learning algorithm.(4)In order to overcome the shortage of comment text and limited context information,a BiLSTM-C algorithm with different granularity is designed.The input of BiLSTM is the fusion of word features and reverse granularity features.Then CNN is used to extract features and classify emotions.At the pooling level,the Max pooling and Avg pooling are carried out respectively.The final features of the text are obtained by cascading the two features through L2 normalization.Finally,the emotional tendency of the comment is obtained through the full connection layer and the Softmax classification layer.The experimental results show that the algorithm can improve the effect of sentiment discrimination in micro-blog comments,and it has certain significance for the research of text sentiment classification.