Research On Sentiment Analysis Based On Deep Learning Model

Emotion analysis is one of the important tasks of network public opinion analysis,which is an important means to guide the trend of network public opinion and maintain social stability.Because language expression has many characteristics,such as semantic ambiguity,metaphorical expression,and the rapid emergence of new words,it brings great difficulties to the task of text emotion analysis.Commonly used machine learning methods have the problem of imperfect text feature extraction,and the emergence of deep learning has brought about a turning point.This paper studies text emotion analysis based on deep learning method.Relevant research contents are as follows:Aiming at the defect that Convolutional Neural Network and Recurrent Neural Network cannot fully extract text emotion features.Combining the improved Capsule network and Attention mechanism,a new emotion analysis model: Capsule-Attention,was proposed.This model uses word2 vec method to vectorize text,extracts text features and highlights important information through Capsule network and Attention mechanism,and outputs binary emotion results using classifiers.Five groups of comparison models were set up,and comparison experiments and ablation experiments were carried out on Yelp,Amazon and SST data sets.Finally,it was proved that the accuracy of this model was greatly improved in the dichotomy of emotion.Aiming at the defect that Capsule Network combined with Attention is difficult to classify multiple emotions,another model: BERT-Bi GRU-Att,is proposed.BERT was used to represent the text vectorization instead of word2 vec,then Bidirectional Gated Recurrent Units was used to extract the feature information of the text,and the Attention mechanism was used to weight the extracted information,finally the five classification results were output.Five groups of comparison models were set up,comparison experiments and ablation experiments were carried out on multiple data sets.It was proved that the accuracy of the model was improved greatly in multiemotion classification.