# OneRel:Joint Entity and Relation Extraction with One Module in One Step

**Authors:** Yu-Ming Shang, Heyan Huang, Xian-Ling Mao  
**arXiv:** [2203.05412](https://arxiv.org/abs/2203.05412)  
**Date:** 2022-03-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.05412v1/

## Key points

- Authors address challenges of joint entity and relation extraction in natural language processing and knowledge graph construction
- Propose a model called OneRel that transforms joint extraction into fine-grained triple classification
- OneRel captures interdependence and indivisibility of the three elements of a triple
- Model consists of scoring-based classifier and relation-specific horns tagging strategy
- OneRel avoids cascading errors and redundant information by treating joint extraction as a single-step process
- Experimental results show that OneRel outperforms existing methods, especially in complex scenarios with overlapping patterns and multiple triples
- Novel perspective on transforming joint extraction into fine-grained triple classification
- Introduction of scoring-based classifier and relation-specific horns tagging strategy for parallel tagging and efficient decoding
- Better performance than state-of-the-art baselines, particularly in complex scenarios involving overlapping triples

## Summary

In this paper, the authors address the challenges of joint entity and relation extraction in natural language processing and knowledge graph construction. They propose a novel model called OneRel that transforms the joint extraction task into a fine-grained triple classification problem. Unlike existing approaches that decompose the task into separate modules or steps, OneRel captures the interdependence and indivisibility of the three elements of a triple. The model consists of a scoring-based classifier and a relation-specific horns tagging strategy. The scoring-based classifier evaluates whether a token pair and a relation belong to a factual triple, while the relation-specific horns tagging strategy ensures an efficient decoding process. By treating joint extraction as a single-step process, OneRel avoids cascading errors and redundant information. The authors evaluate their model on two widely used datasets and compare it with state-of-the-art baselines. The experimental results demonstrate that OneRel outperforms existing methods, particularly in complex scenarios with overlapping patterns and multiple triples. In summary, the contributions of this paper are as follows1. The authors provide a novel perspective by transforming joint extraction into fine-grained triple classification, enabling simultaneous capture of head entities, relations, and tail entities. 2. They introduce a scoring-based classifier and a relation-specific horns tagging strategy to achieve parallel tagging and efficient decoding. 3. The proposed method performs better than state-of-the-art baselines, especially in complex scenarios involving overlapping triples.