Research On Deep Learning-based Sentiment Analysis Method For Comments Text

In recent years,with the widespread popularity of mobile Internet and the hot promotion of various apps,people have started to use various e-commerce platforms and social platforms for their daily consumption and social activities,and post comments on the platforms.User comments are mainly in the form of text and often contain users’ sentiment tendencies and subjective intentions.Analyzing the sentiment tendencies contained in these comments has a positive effect on improving the quality of platform services.Deep learning methods based on neural networks have greatly improved the accuracy and robustness of text sentiment analysis compared with other models,and have become a current research hot topic.However,many CNN-based sentiment analysis studies still suffer from the following problems: the models make insufficient use of features,tend to ignore the contextual dependencies of words,and tend to lose feature information during CNN pooling.In this paper,we propose two sentiment analysis models to address this problem,and the main research is as follows:(1)Proposes an ATT-C-LSTM sentiment analysis model based on the attention mechanism.In order to solve the problem of inadequate feature extraction from Chinese word vectors,the model uses two different methods,cw2 vec and word2 vec,to extract feature information of Chinese words as the input of two channels.One channel uses multiple convolutional windows of different sizes to extract sentiment features of words,thus extracting richer sentiment information of the text;the other channel uses LSTM network to learn the contextual information of the comment text and maintain the sequence relationship between words.The experiments show that the accuracy of ATT-C-LSTM is 92% on the dataset for classification,which effectively solves the problem of inadequate utilization of Chinese feature information in existing models.(2)Propose a MCNN-Caps Net-LSTM sentiment analysis model with multiconvolutional kernels.On the one hand,the model takes into account the advantages of convolutional neural networks in semantic feature extraction and uses multiple convolutional kernels to extract n-gram features of comment text to enhance the feature extraction capability of the model;on the other hand,it draws on the unique feature extraction advantages of capsule networks,which are popular in the field of imagerecognition,and uses capsule networks to solve the problem of easy loss of features in the pooling layer of convolutional neural networks.The advantages of convolutional neural network and capsule network are complemented so that more types of feature information can be extracted.Finally,the model is added to the long and short term memory neural network so that the model can extract both local features and pay attention to the contextual information of the comment text to improve the model’s effectiveness in the sentiment classification analysis task.The experimental results show that dynamic routing of the capsule network can extract features better,and MCNN-Caps Net-LSTM has better performance than other models in the sentiment classification task.