SpanRE: Entities and Overlapping Relations Extraction Based on Spans and Entity Attention

Hao Zhang

2023-04-06

Abstract:Extracting entities and relations is an essential task of information extraction. Triplets extracted from a sentence might overlap with each other. Previous methods either did not address the overlapping issues or solved overlapping issues partially. To tackle triplet overlapping problems completely, firstly we extract candidate subjects with a standard span mechanism. Then we present a labeled span mechanism to extract the objects and relations simultaneously, we use the labeled span mechanism to generate labeled spans whose start and end positions indicate the objects, and whose labels correspond to relations of subject and objects. Besides, we design an entity attention mechanism to enhance the information fusion between subject and sentence during extracting objects and relations. We test our method on two public datasets, our method achieves the best performances on these two datasets.

Computation and Language

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper primarily addresses the problem of extracting triples (entity-relation pairs) in information extraction, especially dealing with the overlap issue among triples. Specifically:

1. **Triple Overlap Problem**:
- The paper defines three types of triple overlap situations: Normal (non-overlapping), SingleEntityOverlap (SEO), and EntityPairOverlap (EPO).
- Previous methods either completely ignored the overlap issue or could only partially address it.

2. **Joint Extraction Model**:
- Traditional information extraction tasks are usually decomposed into two subtasks: entity recognition and relation classification, executed in a pipeline manner.
- This approach overlooks the interrelation between entity recognition and relation classification and is prone to error propagation.
- Joint models can integrate information from both entities and relations, thereby improving performance.

3. **Application of Neural Network Models**:
- Neural network models have achieved success in entity and relation extraction, but constructing elegant architectures to solve the triple overlap problem remains challenging.
- Some previous models, such as Dai et al. (2019), could address partial overlap issues but could not fully solve the EPO problem and had high computational costs.
- Other models, like Takanobu et al. (2019), could handle multiple triples but still extracted inaccurate entity combinations in certain specific cases.

### Main Contributions
- Proposed a novel joint model SpanRE based on span mechanism and entity attention mechanism.
- Designed an innovative label span mechanism that can simultaneously extract objects and relations, fully resolving all possible triple overlap situations.
- Introduced an entity attention mechanism that integrates information from entities and sentences, with experiments showing that this mechanism improves results.
- Tested on two public datasets, with experimental results showing that the model achieved state-of-the-art performance.