Design And Implementation Of Dialogue Emotion Analysis Method Based On Graph Neural Network

Sentiment classification is a special task of text classification.Emotion recognition of dialogue has become a hot research topic in natural language processing.At present,a lot of work has applied convolutional neural network(CNN)and cyclic neural network(RNN)to emotion classification tasks at different levels,but we can't guarantee that CNN and RNN can capture complete context semantic information,so there is still room for improvement in context semantic propagation.For example,in the field of dialogue,only a small part of the work takes the relationship between interlocutors into account in the learning of the model.In order to capture more comprehensive context semantic information and reflect the importance of the relationship between interlocutors to emotion classification tasks,this paper makes a lot of research based on graph neural network,and introduces graph convolution network,graph isomorphic network,graph convolution network and other models to further optimize the problems of context propagation in dialogue.In this paper,a text feature extraction method based on graph convolution network is proposed,and then the final feature representation of dialogue sentences is obtained by combining the pre training model.Finally,the feature representation is used as the input of convolution fusion classification model.The classification model combines two-way longterm and short-term memory network and graph convolution network.The bidirectional long-term and short-term memory network captures the dependency of sequential context and generates semantic representation.The graph convolution network uploads longdistance context information in the dependency of interlocutors.The attention mechanism based on similarity and the attention mechanism based on graph convolution are used to enhance the impact of key semantic features on the classification model.The experimental results show that the feature extraction method based on graph convolution and the convolution fusion classification model have achieved good experimental results,which shows the research value of this paper and the effectiveness of the model.