Sentiment Analysis Of Chinese Book Reviews Based On Sentiment Lexicon And Deep Learning

With the continuous development of science and technology,digital information brings great value to people,which can not only help businesses find the shortage of goods,but also help consumers choose their favorite goods.When the amount of data is increasing,the traditional sentiment analysis method can't meet the needs of massive data processing,and the effect is too dependent on artificial features.The thesis takes the mainstream research methods of sentiment analysis and BERT pre-training model as the background,and combines sentiment lexicon and deep learning to conduct sentiment analysis of Chinese texts.Firstly,BERT word vector and sentiment lexicon are combined as the feature matrix of sentences.Secondly,onedimensional convolution layer is used to extract the features of text information,compress the dimensions of word vectors,and then the main sentiment features and contextual features in comments are obtained by CNN and Bi GRU network.Finally,the weighted sentiment features are classified by attention mechanism,and compared with other model results.According to the experimental results,the effect of word vector splicing emotion dictionary proposed in this paper is better than that of word vector weighting by sentiment lexicon.Moreover,the one-dimensional convolution layer added before CNN layers can play a very good role in learning word vector features.The accuracy and F1 score of the final model are 94.67% and 94.65%,respectively.Compared with SLCABG model,the effect is improved by more than 1%.