Emotion Analysis Model Based on Multi-head Attention and Tree_LSTM
Zhengjun Pan, Lianfen Zhao, Yang-hong Mao · 2022
Most of the emotion analysis models based on attention mechanism and neural network adopt single-layer attention mechanism, and the feature expression is relatively single, which can not be applied to various types of sentences. Based on the existing attention mechanism and neural network model, this paper proposes a method to fuse emotional information and combine multi head attention mechanism and Tree_LSTM's emotion analysis model. aims to enable the model to obtain more levels of information about sentences in different representation spaces, dynamically adjust the feature weight through the multi head self attention mechanism, and introduce emotional information on the basis of existing network input to further obtain more text syntax information and improve the feature expression ability of the model. Finally, the emotion category is obtained by softmax classifier. The experimental results of coae2014 microblog data set show that compared with Bi_LSTM, CNN_LSTM, Self-attention_BILSTM, Self_attention_Tree_BiLSTM model, which further improves the performance of emotion classification.