Enhanced semantic natural language inference algorithm based on BERT-CNN

• 1、School of Computer Science, Beijing University of Posts and Telecommunications, Beijing,

Abstract：Natural language inference is a basic task in current natural language processing technology, which can be used to evaluate the model\'s ability to understand semantics between texts. In view of the insufficient understanding of the semantics between text pairs in current research, this paper proposes a natural language reasoning model based on BERT-CNN with enhanced semantics. The model first extracts the feature vector of the overall interactive classification of the text through the basic BERT network and the semantic vector of each character of the two texts, and then uses the CNN network to encode the semantic vector of each character of the two texts to obtain the features of the two texts Vector, and finally concatenate the feature vector of the overall interactive classification and the feature vector of the two texts and use the forward network for classification prediction. The experimental results show that, compared with the traditional natural language inference algorithm, the enhanced semantic natural language inference algorithm based on BERT-CNN proposed in this paper has higher accuracy and further improves the performance of the model, thus verifying the effectiveness of the proposed method

Keywords： natural language inference BERT CNN