Research On Text Sentiment Analysis Method Based On BERT And Hybrid Neural Network

As an important research direction in the field of natural language processing,text sentiment analysis can effectively analyze all kinds of sentiment information contained in the text.In the face of the explosive growth of Internet text resources,it is of great significance to make effective use of text data and explore the commercial value and research value behind it.In the era of big data,the replacement and updating speed of network words is faster,and the method based on sentiment dictionary needs a lot of manpower and financial resources to constantly update the sentiment dictionary constantly;the method based on Machine Learning depends on manual annotation of text,so it is difficult to learn deeper text semantic information.In this situation,the method based on Deep Learning has emerged,which can easily handle massive data and automatically extract features,and has strong generalization ability.Convolutional Neural Networks and Recurrent Neural Networks and its variants are widely concerned in the field of natural language processing with their respective advantages.However,the current Deep Learning methods also have many problems in the process of dealing with text information,such as the low classification accuracy of a single network model,the same attention to all words and the inability to solve the polysemy of words.In view of these problems,the main research work of this paper is as follows:(1)In order to solve the problems such as the disappearance of gradient in RNN and the long-term dependence caused by being unable to master the nonlinear relationship with long time span,this paper uses the Bi LSTM network for serialization learning,which not only solves the problem of gradient disappearance of RNN,but also pays better attention to context information when analyzing text.Combined with the advantage that CNN can extract local features and semantic information of text,a text emotion analysis method based on CNN-Bi LSTM model is proposed.The experimental results show that the CNN-Bi LSTM model has a better effect of sentiment analysis than the traditional single CNN or Bi LSTM model.(2)the attention mechanism can give different weights to words with different importance and Text-CNN can extract different sentiment features with different granularity and the relationship between sentences by using convolution kernels of different sizes,as well as the advantages of BiGRU in dealing with dependency information between words with large distances,which is further optimized on the basis of CNN-Bi LSTM model.In this paper,a text sentiment analysis method based on the attention mechanism of parallel hybrid network is constructed.Firstly,the Glove model is used to train the text to get the word vector,then Text-CNN is used to extract the sentiment features of different granularity,BiGRU is used to obtain the dependency relationship between words with large time step distance,and then the extracted information is fused,and then pay close attention to the features of important words through the attention mechanism,which further improves the accuracy of classification and has strong application ability.(3)the traditional language model and word vector representation methods such as Word2 vec and Glove can not solve the problem of word polysemy.To solve this problem,this paper uses BERT pre-training model to dynamically obtain word vectors as the input of Machine Learning and Deep Learning algorithms for sentiment feature extraction and sentiment polarity classification.The open data set is used for experiments to verify the effectiveness of the BERT model,and then the model is applied to the mental health analysis text data.The experimental results fully prove the validity and universality of the model.