Research On Text Emotion Analysis Based On Deep Learning

With the rapid development of the Internet in the 20 th century and the rapid growth of the scale of Internet users,people began to get used to communicating online and expressing their views on something.Behind this information,there are huge commercial and social values.Emotion analysis is a kind of data analysis and processing for the emotional tendency of these online comments,which has become an important task in industry and academia.However,at present,there are still many problems in emotion analysis,such as insufficient learning of key information and insufficient accuracy of emotion classification.In view of the weak ability of extracting key information in current emotion analysis models,they cannot effectively capture the internal correlation between features,thus ignoring the impact of non-key information on the results,this chapter presents a WT-SG-M model incorporating self-attention mechanism.The model combines Word2 vec with the term frequency-inverse document frequency(TFIDF)algorithm,and adds a self-attentional mechanism to the gated recurrent unit(GRU)network model to highlight the role of keywords in the text through weighted probability.At the same time,the Maxout neuron was introduced into the output of the model,which effectively alleviated the gradient dispersion problem.In view of the problems of long training time and insufficient information learning in the following text emotion analysis,this paper proposes a hybrid neural network combined with conditional random field to improve the model H-CRF.Firstly,a text is divided into different regions according to the sentence,and the semantic information and feature vectors output by two improved neural network models,CNN and GRU,are combined.Then,the vector representation of sentences through different network models is spliced together,and the conditional random field model is used as a classifier to ensure the accuracy of emotion analysis task on the basis of effectively reducing training time.Through several groups of analysis and comparison experiments,the emotional classification effect of the improved model was verified.Through the comparison and analysis of experiments,it is concluded that the method proposed in this paper has improved the evaluation indexes to some extent compared with several models commonly used at present.The model in this paper has more sufficient features,faster training speed and better classification effect than previous models.