Research on Sentence Similarity Calculation Based on Attention Mechanism and Sememe Information

Abstract

Focusing on the research of sentence similarity calculation, this paper proposes a method combining bidirectional long short-term memory networks, attention mechanism and sememe (BILSTM-ATTENTION-SEMEME) to achieve better results on semantic representation and in-depth understanding of the semantic level, consequently better resolve the problem in the aspect of semantics in the field of intelligent customer service. This method first solves the semantic representation problem through a model based on bidirectional long short-term memory networks and attention mechanism (Bilstm-Attention), then combines the sememe information of HowNet in the training of word vectors to improve the performance of semantic under-standing. Experimental results show that the proposed method is effective in the computation of sentence similarity in the field of intelligent customer service, and it can well combine the sememe knowledge of HowNet with the deep learning model based on attention mechanism. Compared with the baseline system, the accuracy rate increased by 6.5%.