Research On Emotion Classification Of Texts Based On Deep Learning

With the increase of text information on the Internet,mining the potential value behind massive text information has become an important demand for people,and text classification is the key technology among them..As a branch of text classification task,text emotion classification has important research value and a wide range of application scenarios.In recent years,deep learning methods have been applied to text emotion classification tasks.Although deep learning methods have achieved good results,their accuracy still needs to be improved.This paper proposes new deep learning text emotion classification models based on attention mechanism and multi-head mechanism,which alleviate the singleness problem caused by the use of maximum pooling,and improve the effect of text emotion classification.The specific work is as follows.Firstly,We introduced an attention mechanism to improve the pooling layer of the classic recurrent convolutional neural network model,and designed a recurrent convolutional neural network model based on attention pooling.This model uses attention pooling instead of maximum pooling.In the pooling process,instead of focusing on the most prominent feature information,different weights are assigned according to the contribution of each feature information to the classification,so as to achieve the goal of full attention to all feature information.This method can effectively balance comprehensiveness and pertinence.Secondly,We introduced a multi-head mechanism,and designed a recurrent convolutional neural network model based on multi-head maximum pooling and a recurrent convolutional neural network model based on multi-head attention pooling.The recurrent convolutional neural network model based on multi-head maximum pooling performs multiple different projection transformations for the input of the pooling layer.Then the maximum pooling operation is performed for different projection transformation results to alleviate the original problem of singleness caused by the single maximum pooling.The recurrent convolutional neural network model based on multi-head attention pooling performs multiple different projection transformations for the input of the pooling layer,and each head performs different attention pooling operations.This model combines the advantages of the attention mechanism and the multi-head mechanism,and can extract text feature information more comprehensively and fully.Thirdly,we conducted a text emotion classification experiment.We conducted sufficient experiments on the emotion classification data set with the model mentioned in this article and multiple comparison models,and conducted evaluation and analysis through various indicators.The experimental results show that the models proposed in this paper effectively improve the effect of text emotion classification,and the multi-head attention pooling recurrent convolutional neural network model performs best.