Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Decoupling and Aggregating Framework for Joint Extraction of Entities and Relations

Yao Wang, Xin Liu, Weikun Kong, Hai-Tao Yu, Teeradaj Racharak, Kyoung-Sook Kim, Minh Le Nguyen

TL;DR

The paper tackles joint extraction of entities and relations by addressing limitations of shared-feature methods and coarse information interaction. It introduces DArtER, which decouples encoding into three subtask-specific streams (ES, EO, ER) and employs inter- and intra-aggregation to build fine-grained representations, with a BiDArtER extension for bidirectional context; decoders then fuse these features to predict NER spans and relation triples. Extensive experiments on seven benchmarks (ACE04/05, CoNLL04, ADE, SciERC, NYT, WebNLG) show state-of-the-art or competitive results, with ablations confirming the value of fine-grained encoding and task-interaction strategies. The work demonstrates that richer, subtask-specific feature construction and cross-subtask interaction can significantly improve joint NER and RE, especially for IT sentences, and highlights areas for handling long-tail entities and complex domains.

Abstract

Named Entity Recognition and Relation Extraction are two crucial and challenging subtasks in the field of Information Extraction. Despite the successes achieved by the traditional approaches, fundamental research questions remain open. First, most recent studies use parameter sharing for a single subtask or shared features for both two subtasks, ignoring their semantic differences. Second, information interaction mainly focuses on the two subtasks, leaving the fine-grained informtion interaction among the subtask-specific features of encoding subjects, relations, and objects unexplored. Motivated by the aforementioned limitations, we propose a novel model to jointly extract entities and relations. The main novelties are as follows: (1) We propose to decouple the feature encoding process into three parts, namely encoding subjects, encoding objects, and encoding relations. Thanks to this, we are able to use fine-grained subtask-specific features. (2) We propose novel inter-aggregation and intra-aggregation strategies to enhance the information interaction and construct individual fine-grained subtask-specific features, respectively. The experimental results demonstrate that our model outperforms several previous state-of-the-art models. Extensive additional experiments further confirm the effectiveness of our model.

A Decoupling and Aggregating Framework for Joint Extraction of Entities and Relations

TL;DR

The paper tackles joint extraction of entities and relations by addressing limitations of shared-feature methods and coarse information interaction. It introduces DArtER, which decouples encoding into three subtask-specific streams (ES, EO, ER) and employs inter- and intra-aggregation to build fine-grained representations, with a BiDArtER extension for bidirectional context; decoders then fuse these features to predict NER spans and relation triples. Extensive experiments on seven benchmarks (ACE04/05, CoNLL04, ADE, SciERC, NYT, WebNLG) show state-of-the-art or competitive results, with ablations confirming the value of fine-grained encoding and task-interaction strategies. The work demonstrates that richer, subtask-specific feature construction and cross-subtask interaction can significantly improve joint NER and RE, especially for IT sentences, and highlights areas for handling long-tail entities and complex domains.

Abstract

Named Entity Recognition and Relation Extraction are two crucial and challenging subtasks in the field of Information Extraction. Despite the successes achieved by the traditional approaches, fundamental research questions remain open. First, most recent studies use parameter sharing for a single subtask or shared features for both two subtasks, ignoring their semantic differences. Second, information interaction mainly focuses on the two subtasks, leaving the fine-grained informtion interaction among the subtask-specific features of encoding subjects, relations, and objects unexplored. Motivated by the aforementioned limitations, we propose a novel model to jointly extract entities and relations. The main novelties are as follows: (1) We propose to decouple the feature encoding process into three parts, namely encoding subjects, encoding objects, and encoding relations. Thanks to this, we are able to use fine-grained subtask-specific features. (2) We propose novel inter-aggregation and intra-aggregation strategies to enhance the information interaction and construct individual fine-grained subtask-specific features, respectively. The experimental results demonstrate that our model outperforms several previous state-of-the-art models. Extensive additional experiments further confirm the effectiveness of our model.
Paper Structure (28 sections, 10 equations, 3 figures, 15 tables)

This paper contains 28 sections, 10 equations, 3 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Definition
  4. Methodology
  5. Encoder
  6. Decoder
  7. BiDArtER
  8. Training
  9. Experiments
  10. Datasets
  11. Experimental Settings
  12. Baseline Models
  13. Results and Analysis
  14. Ablation Study
  15. Number of the DAM encoder layers
  16. ...and 13 more sections

Figures (3)

  • Figure 1: The overall framework of the DArtER model.
  • Figure 2: The overall framework of the BiDArtER model.
  • Figure 3: Comparison of different relational types on ACE2004, ACE2005, and NYT datasets. The scores reported for the PFN model are re-implemented.