Attention-based ALBERT-BiLSTM-CNN for sentiment classification of comments
In the field of natural language processing,the Word2Vec representation method for static word vectors is unable to accurately capture the semantic information of sentences and text structures.Based on this problem,we propose a comment senti-ment analysis model based on ALBERT-CNN-BiLSTM-Attention,which not only effectively solves the problem of high number of parameters and low computational and storage efficiency of the traditional model,but also significantly improves the accuracy of sen-timent classification.The method firstly utilizes ALBERT model to obtain dynamic features of comment text,and introduces the at-tention mechanism module to obtain text keyword weights for the output results of BiLSTM,and then utilizes CNN to obtain local fea-tures of the text,and finally carries out sentiment classification of the comment content through Softmax layer.The experimental re-sults show that the model in this paper has a significant improvement in precision and recall compared with the traditional methods and other models of ALBERT,with the precision reaching 88.43%,the recall reaching 88.17%,and the F1 value reaching 88.30%.
NETL
NSTL
万方数据
