Dual-channel attention convolutional neural network sentiment analysis model fusing strokes and primitives

Aiming at the problems that a traditional deep learning method uses a single word vector to cause insufficient representation capability and the prior art neglects internal information of Chinese vocabularies, the invention aims to introduce more prior knowledge of Chinese languages so as to have stronger representation capability on Chinese texts and finally, the accuracy of the Chinese sentiment analysis task is improved. The invention provides a dual-channel convolutional neural network (SS-DCCNN) model fusing stroke and primitive features, and semantic information for Chinese vocabularies is better fused. An attention mechanism is used to identify different influences of each word in a sentence on a specific task classification result, so that words with relatively large relation coefficients with other words in the sentence are more concerned. Specifically, morphological characteristics of Chinese characters are introduced by using a cw2vec method, HowNet knowledge base information is extracted by using