The Study on Relational Triplet Extraction Method Based on Dependency Structure
Information extraction is the key task in natural language processing,and relational triplet extraction is the core subtask of information extraction.In the relational triplet extraction task,using the dependency structure information of the sentence can strengthen the global understanding of the sentence,and can help improve the extraction effect of the model.The extraction method based on dependency structure analysis and graph neural networks is proposed.Firstly,the model uses the pre-trained model to obtain the semantic representation of the text vector.Secondly,the model obtains the dependency structure information of text and constructs the graph,and uses the graph neural network to obtain the global understanding of the structure information of the graph.Finally,the model extracts the relational triples contained in the text through a specific decoding method.The experimental results show that compared with the existing combined extraction models on NYT29,NYT24 and WebNLG,the accuracy of the proposed method is improved by 0.1％～0.6％,the recall rate is improved by 0.2％～0.5％,and the F1 value is improved by 0.1％～0.3％.
information extractionrelational triplesdependency structure analysisgraph neural network
万方数据
维普
