Research On Sentiment Analysis Of Comment Text Based On Hybrid Neural Network

Network information technology develops rapidly,the network platform has gradually become the main place for people to express their personal comments and opinions,and most of them are presented in the form of text,which makes the task of text sentiment analysis attract the attention of the majority of researchers.In the past,the main research methods of text sentiment analysis are divided into sentiment lexics-based and machine learning.The construction process of sentiment dictionary is time-consuming and time-consuming,and the effect of sentiment classification depends on the quality of sentiment dictionary,which makes the universality of sentiment dictionary in different task fields poor.Deep learning model has become the mainstream research method by virtue of automatic feature learning.However,the feature extraction ability of the basic deep learning model is weak,and the semantic representation quality of the static word vector model is poor,which leads to the ambiguity and incompleteness of the feature representation.Starting from the task of text sentiment analysis,this thesis constructs on the basis of traditional deep learning models:the model combining BERT and Bi SRU-Attention,and the sentiment analysis model combining ALBERT and hybrid network as the main research algorithms.The main contents of this thesis are as follows:(1)This thesis proposed a text sentiment analysis model combining BERT and Bi SRU-Attention,and used the pre-trained language model combined with the specific context of the current word to adjust the dynamic vector representation.The Bi SRU module maintains an efficient understanding of sentences,and the soft attention mechanism identifies key emotional words.Problems such as the inability of traditional word vectors to distinguish polysemy and the lack of feature extraction ability of the basic deep learning model are improved.Experiments are conducted to verify the improvement of the recognition performance of the model on text emotion.(2)This thesis proposes a text sentiment classification algorithm combining ALBERT and hybrid network,which combines the ordered neuron LSTM with soft attention mechanism and multi-scale convolutional neural network to comprehensively capture the sentiment features of different aspects of comment text.The traditional loop module can only learn the information features of text sequence,and lacks the capture of sentence hierarchy information and local emotion features,so as to enhance the understanding ability of the model on emotion semantics.Figure [35] Table [13] Reference [66]...