Joint Extraction Of Chinese Entity Relationship Based On Bidirectional Semantic Learning Model

Entity relationship extraction aims to extract structured data with rich semantic information from unstructured text,It’s an important research task in the field of natural language processing.However,Chinese named entity relationship extraction has some unique linguistic characteristics,such as the complexity of Chinese language structure and the lack of obvious segmentation boundaries,which have been seen as the main challenges to develop a high performance of a Chinese entity relationship extraction model.This thesis presents the use of bidirectional semantic collaboration to boost the effectiveness of entity relationship extraction.On this foundation,a relation extraction algorithm based on bidirectional relational link is further proposed to extract overlapping relation triples.The previous entity relationship extraction methods usually only use the one-way semantic features in the text for extraction.The effectiveness of this method needs to be improved.Therefore,we propose a joint extraction model of entity relationship triples based on two-way semantics for restricted Chinese entity relationship extraction.The approach relies on a Ro BERTa pretrained language model to obtain text representations with contextual information,then identifies entities in sentences in the form of pointer annotations.We also designed a positive and negative relationship mapper based on bidirectional semantics.Relation triple candidates with the corresponding probabilities were then generated from the perspective of positive and negative relationships.We chose the best relation triple based on those probabilities.Experiments show that the model improves the F1 value by an average of 12.8%,which boosts the effectiveness of Chinese entity relationship extraction.Aiming at the issue that traditional methods cannot effectively extract overlapping relation triples in open entity relationship extraction,we further propose an entity relationship extraction model based on bidirectional relational links.By defining bidirectional relational links,the model transforms the task of entity relation extraction into identifying bidirectional relational links in sentences,thereby effectively extracting overlapping relation triples in sentences.The model first uses the NEZHA pre-trained language model to obtain text representations,then simultaneously recognizes entities and relational words in the sentence based on the global pointer.After obtaining the entities and relational words in the sentence,it applies the bidirectional relational link discrimination module to estimate whether there is a positive or negative relational link between each pair of entities and relational words.Finally,the obtained bidirectional relational links were combined to derive relation triples.Experiments show that the model achieves the highest F1 value compared with other entity relationship extraction methods,which is 76.9%.It is proved that the model can effectively extract overlapping relation triples.