Research On Chinese Textual Entailment Recognition Method Incorporating Knowledge Graph Embedding And Self-attention

Text entailment recognition is an important task in the field of natural language processing.The purpose of this task is to judge whether there is an entailment relationship between two sentences,that is,to judge whether a sentence can deduce another sentence.entailment relationship is widely distributed in natural language texts,and it is a one-way reasoning relationship.In recent years,the Transformer model based on the self-attention mechanism brought a huge impact on the field of natural language processing.This model has also been applied in text entailment recognition tasks and achieved excellent results.However,the Transformer model is still inadequate in modeling local information of text.Moreover,most of the existing text entailment models are based on word vectors,which fail to make full use of external knowledge resources,but judging entailment often requires some knowledge other than word vectors.Therefore,this paper focuses on improving the Transformer model and introducing external knowledge in the Chinese text entailment recognition task,and mainly does the following work:To overcome the deficiency of Transformer model in text local information modeling,this paper introduces a Quasi-Recurrent Neural Network in Transformer model.On the one hand,before attention calculation,the input text sequence is divided into local short sequences using QRNN network to capture the local information of the input text.On the other hand,combining the gating mechanism improves self-attention so that the model chooses tasks-related words or features.The improved Transformer model can effectively raise the accuracy of entailment relationship recognition.In order to introduce external knowledge resources into the text entailment model,this paper uses the knowledge graph embedding method to introduce the common sense knowledge in the Concept Net to enrich the semantics of the input text in the text entailment model.Experiments show that the introduction of external knowledge can effectively improve the recognition performance of the text entailment model on smallscale datasets.In addition,this paper also explores the application method of text entailment technology in text summarization system,by using text entailment model to rerank multiple candidate summaries to reduce factual errors in generating summaries.