A Novel Sentiment Classification Architecture Based on Self-attention Mechanism

Abstract

Due to the huge amount of network data and low efficiency of artificial analysis, one branch of natural language processing, sentiment classification, has become the focus of current research. Attention mechanism has been widely applied in the field of natural language processing, and the accuracy is higher, but due to the different aspects in a sentence to express sentiments tend to be different and attention mechanism is difficult to focus on more text sequences of internal relations, this paper studies the part-of-speech embedded into the attention mechanism, based on fusion embedded speech sentiment classification new architecture since the attention mechanism, including the two kinds of algorithms, respectively is based on the Pos-IdSA sentiment classification algorithm and based on the Pos-ItSA sentiment classification algorithm; The parameters of the data set are optimized and the performance of the algorithm is analyzed. The research in this paper shows that the new algorithm of sentiment classification based on fused part-of-speech embedding has a higher accuracy rate, and part-of speech embedding is conducive to learning the grammatical relationship between different parts of speech on the network, and effectively reduces the noise caused by the complex context.