A Two-stage Chinese Text Emotion Recognition And Application Research Based On Deep Learning And Dictionary

Nowadays,people constantly express opinions and comments on commodities,person and social events on the Internet platform.The analysis and mining of these massive comment data is of great practical significance Therefore,the research on the automatic classification of network text data,especially the automatic recognition of emotion in the text,has become more and more importantThis work mainly studied the emotion recognition of Chinese network opinion text and the digital measurement of text emotion intensity.The process mainly included the following works:Firstly,the crawler and other methods were used to obtain the real network public opinion corpus.Then,the corpus is preprocessed and various semantic dictionaries were constructed.After that,the Word2Vec was used to realize the vectorization of corpus.Furthermore,the traditional machine learning-based methods(Support Vector Machine and Naive Bayes),the deep learning-based methods(Recurrent Neural Network,Convolutional Neural Network and their variants),and the semantic dictionary-based methods were used to realize the emotion recognition task,and these three methods were optimized accordingly.In addition,the approach of feature selection and classifier selection were studied during the machine learning-based methods;the way of calculating the similarity of semantic words and automatic extending the domain emotion dictionary based on semantics were also studied during semantic dictionary-based methods.Finally,a two-stage text emotion recognition model which based on deep learning and semantic dictionary is proposed in the present study to address the defects of these two methods,and the performance of models were evaluated according to recall rate,accuracy rate and F1value.Experiments show that the proposed two-stage recognition model combines almost all the advantages of the above methods,which can not only maintain the highest recognition performance,but also realize the digital measurement of text emotion.In order to make the research results can be better implemented into practical applications,this study finally developed a network public opinion recognition system using the python programming language and PyQt5 technology in the Spyder integrated development environment.The system can automatically complete the emotion recognition of network opinion text,and output the emotion category,emotion intensity score,key words or sencentences,word cloud map and other information.In addition,users can also use their own corpus to retrain the model to improve the accuracy of the recognizer.