Abstract:

A new two-channel sentiment analysis method,C-A-BiLSTM,is proposed to solve the problems that the traditional sentiment analysis method has a low accuracy and cannot fully extract text feature information.The model performs convolution operations on two different channels in different directions of word vectors and Word-POS word vectors to mine deeper semantic information,in which the word vector channel extracts more semantic local information and effectively alleviates the problem of unlisted words in the thesaurus.The word vector channel uses the part of speech tagging technology to obtain the part of speech of the corresponding word,which solves the problem of polysemy of one word faced by the original word vector.The combination of the two channels can efficiently mine deeper semantic and grammatical information,but it is unable to filter the key information from the text tensor,which consumes a lot of computational power.Therefore,the attention mechanism is introduced,on the basis of which the A-BiLSTM network combined with the Attention mechanism is used to further extract context information and to gain more comprehensive and high-quality features.Experimental achievements indicate that the accuracy,recall and F1 values which the model has reached all exceed 94%,which is notably enhanced in comparison with the CNN algorithm,SVM and BiLSTM algorithm,and that the error rate is reduced by about 1%~6%.The method has a certain advantage in text analysis tasks.

Key words: CNN, BiLSTM, text sentiment analysis, word vector, Word-POS vector