Text Sentiment Classification Based On Deep Learning And Attention Mechanism

The purpose of natural language processing is to use computer technology to analyze human-generated text data and to understand human intentions.The ultimate goal is to enable unobstructed communication between human language and computers.It mainly includes fields such as machine translation,sentiment analysis and intelligent question answering.The main task of text sentiment analysis is to mine and analyze texts with subjective emotions to derive their emotional tendencies.With the rapid development of the Internet,the implementation of deep learning algorithms has been supported by more solid technology,which has made great contributions in various fields.Using deep learning methods to solve the problem of natural language processing has become the choice of more researchers.Based on the background of text sentiment analysis,this paper summarizes relevant research methods and techniques,and studies commonly used algorithm models,such as convolutional neural networks and recurrent neural network.In addition,The improved algorithm models are proposed,which have certain effects in improving the accuracy.This paper first discusses the related theoretical knowledge of word vector technology,traditional machine learning algorithm,deep learning algorithm and attention mechanism,and introduces the working process of convolutional neural network,recurrent neural network and attention mechanism.Then the steps and methods of training word vectors using Word2 vec model are introduced.In order to further improve the quality of Chinese word vectors,up to 1G corpus related to data set field is used for training,and the specific models and parameters in the training process are explained.Then a hybrid text sentiment analysis model CNN-GRU-Attention is proposed to solve the problem of excessive feature loss in the pooling layer of traditional convolutional neural networks.The model uses GRU instead of the traditional pooling layer of convolutional neural network,and introduces attention mechanism to emphasize the influence of important information features.It is evaluated by Chinese comments collected on the Internet.The experimental results show that the model can improve the accuracy of text classification.Finally,aiming at the insufficient feature extraction of single channel network with the increase of layers,a two-channel attention text sentiment analysis model is proposed.The word vector matrix is used as the input of the convolution network channel and the long and short memory network channel to learn the local features and sequence features of the text respectively.In addition,the attention layer is used to learn important features to obtain the probability distribution of text feature importance.The model is trained using Jingdong Shopping Review and comparative experiments are conducted to verify the results,which show the feasibility and effectiveness of the two-channel model.