Research And Application Of PL&A Entity-Relationship Model Oriented To Text

Information extraction is one of the key technologies in the field of natural language processing,which usually refers to extracting useful information from a large amount of data.Unstructured data on the Internet is mostly irregular,multi-type,and irregular text data.Traditional entity recognition and relationship extraction are mostly two independent task forms.The extraction method is to separate the entity extraction and the relationship extraction.This extraction method is easy to cause many problems such as entity redundancy,error transmission,and relationship overlap.In this thesis,aiming at the problem of pipeline,we study the PL&A entity-relationship joint extraction model based on Pointer Labeling and Attention Mechanism,and apply the model to construction and application of ancient poetry knowledge graph.The main research work of this thesis includes the following aspects:(1)To improve the joint extraction model,the dynamic word embedding method is adopted in the input stage,which is different from the traditional entity relationship extraction idea,and the parent entity and relationship are used to predict child entities,and the relationship type is no longer regarded as a discrete label,the relationship extraction stage uses the form of multiple pointer annotations for annotation.And the above two tasks are modeled in a unified way to achieve parameter sharing between the two tasks and improve the performance of the joint extraction of the model.Finally,the effect of the model is verified by comparative experiments,ablation experiments and training parameter comparison experiments.Experiments prove that the model has achieved good results in identifying overlapping relations of text and solving entity redundancy.(2)After determining the demand for ancient poetry knowledge services,carry out intelligent service research based on ancient poetry knowledge graphs.Apply the PL&A entity-relationship joint extraction model to the crawled ancient poetry data.According to the characteristics of Chinese text data in terms of word embedding,word fusion is adopted in word embedding,and the form of word vector and word vector splicing is innovated to improve the representation ability of input sentences.And use the manually-labeled ancient poetry dataset to train the entity-relationship joint extraction model to build a knowledge map based on the ancient poetry text data.Finally,based on the ancient poetry knowledge map and joint extraction model,the application of ancient poetry knowledge such as visual query,intelligent question answering and entity relationship extraction is designed.