Research And System Implementation Of Entity Relation Joint Extraction Based On Deep Learning

Entity relation joint extraction aims to automatically extract structured triplets from unstructured text,which is of great significance for the construction of knowledge graphs and semantic question answering.However,existing joint extraction models mainly address the issues of entity overlap and relation overlap,neglecting the error propagation caused by entity recognition,which leads to the incorrect placement of entities in relation extraction and reduces the accuracy of entity relation extraction.In addition,the importance of word part-of-speech features and dependency relationship features in text has been neglected in existing models.To address these issues,this paper designs a part-of-speech attention mechanism and a classifier that integrates part-of-speech and dependency relationship features to improve entity relation extraction accuracy.The main research work of this paper is as follows:(1)To address the error propagation caused by entity recognition,a part-of-speech attention mechanism is designed based on the part-of-speech and dependency relationship features of words in a sentence.This mechanism optimizes the word representation obtained by a pre-trained language model by incorporating part-of-speech information into the word representation.Prior knowledge such as part-of-speech,dependency relationship,and entity length are added to the entity classifier and relation classifier.The experimental results on public datasets show that the entity relation classifier that integrates prior knowledge such as part-of-speech can significantly improve classification accuracy,and the part-of-speech attention mechanism module works in collaboration to improve the accuracy of the entity relation extraction model.(2)To address the problem of entity overlap and relation overlap in joint extraction models,a fragment annotation strategy is adopted,and entity and relation classifiers are designed based on the word representation obtained by a pre-trained model.In the entity classifier,the problem of nested entities is solved by enumerating all candidate entity fragments.In the relation classifier,the problem of entity overlap and relation overlap is solved by enumerating all candidate entity pairs and predicting the relationship between them.(3)A digital library system for universities is designed and implemented.The text triplet extraction method is applied to extract important knowledge from university textbooks,and the obtained triplets are used to construct knowledge graphs and realize functions such as knowledge retrieval,relationship retrieval,and semantic question answering.This system solves the problems of complex knowledge relationships and multiple course contents in actual educational scenarios and improves the learning efficiency of learners.