Research On Short Text Emotion Analysis Based On CNN And RNN

With the development of social networking,e-commerce,mobile Internet and other technologies,various network data has rapidly increased,and the Internet contains a large amount of emotional text data.How to automatically analyze and process short text from different channels has become an urgent problem to be solved.Emotion analysis belongs to a branch of natural language processing.In recent years,many scholars have studied it.Research on emotion analysis of short text based on CNN and RNN is to analyze and mine short text information such as microblog comments and shopping evaluation on the Internet through CNN and RNN related algorithms,and to identify whether the text contains emotions,positive and negative emotions and emotion categories.The main work of this article includes the following four aspects:First,a DB-AC model is proposed for the emotion analysis task of Weibo text.Explored the role of emotion dictionary and sample balance module in Weibo emotion analysis task,and found that the introduction of emotion dictionary can improve the effect of emotion classification,and the fine-grained emotion dictionary is obviously superior to the traditional emotion dictionary;For the balance problem,a sample balance module combining oversampling and under-sampling is constructed.A DB-AC model that combines a fine-grained emotion dictionary and a sample balance module is proposed.It achieves good performance in the emotion classification task of Chinese microblog emotion analysis,and it also improves emotion recognition and sentiment classification tasks.Secondly,a character and word fusion model is proposed for sentiment classification and emotion recognition.Short text on the Internet is more serious in terms of irregular language,and the word segmentation effect is relatively poor.Characters and words have their own advantages as basic units of deep learning models,but they do not take into account the correlation between characters and words.This paper proposes a new character and word fusion model based on the bidirectional Long and Short-Term Memory(Bi LSTM)and the Convolutional Neural Network(CNN)models,which can fully consider the semantic information between characters and words.Experimental results show that this method can improve the effects of sentiment classification and emotion recognition.Then,a two-channel Bi GRU-CNN-Attention model is proposed for sentiment classification and emotion recognition of Weibo text.The traditional single-channel and single-model research methods are difficult to obtain the deep semantic and context information in the text at the same time.In order to obtain better classification results,the bidirectional Gated Recurrent Unit(Bi GRU)and CNN models are combined to construct a parallel dual-channel Bi GRU-CNN model,which can achieve better results than the single Bi GRU and CNN models.On this basis,the attention mechanism is introduced to construct the Bi GRU-CNN-Attention model,which verifies that the self-attention mechanism can screen typical features and further improve the model effect.Finally,a CW?BGCA model is proposed for sentiment classification of shopping reviews.The design implements the character level C?BGCA and the word level W?BGCA models.Both models use CNN to learn the context features extracted by Bi GRU,and add attention mechanisms to form a hybrid neural network,and finally form a two-channel hybrid neural network model(CW?BGCA).This proves the effectiveness of the dual-channel hybrid neural network in the task of analyzing emotion in shopping reviews.