Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explore BiLSTM-CRF-Based Models for Open Relation Extraction

Tao Ni, Qing Wang, Gabriela Ferraro

TL;DR

The paper tackles open relation extraction in multi-relation sentences by deploying BiLSTM-CRF models with both static and contextualized embeddings and introducing a new tagging scheme to resolve overlapping relations. It demonstrates that an orderly, multi-relational tagging approach, especially when paired with BERT-based embeddings, yields high recall and Predicate Matching Scores, surpassing traditional baselines. The work provides a practical advancement in multi-relational extraction and outlines future directions for joint extraction and cross-lingual applications. This approach enhances the robustness and scalability of Open RE systems in real-world text.

Abstract

Extracting multiple relations from text sentences is still a challenge for current Open Relation Extraction (Open RE) tasks. In this paper, we develop several Open RE models based on the bidirectional LSTM-CRF (BiLSTM-CRF) neural network and different contextualized word embedding methods. We also propose a new tagging scheme to solve overlapping problems and enhance models' performance. From the evaluation results and comparisons between models, we select the best combination of tagging scheme, word embedder, and BiLSTM-CRF network to achieve an Open RE model with a remarkable extracting ability on multiple-relation sentences.

Explore BiLSTM-CRF-Based Models for Open Relation Extraction

TL;DR

The paper tackles open relation extraction in multi-relation sentences by deploying BiLSTM-CRF models with both static and contextualized embeddings and introducing a new tagging scheme to resolve overlapping relations. It demonstrates that an orderly, multi-relational tagging approach, especially when paired with BERT-based embeddings, yields high recall and Predicate Matching Scores, surpassing traditional baselines. The work provides a practical advancement in multi-relational extraction and outlines future directions for joint extraction and cross-lingual applications. This approach enhances the robustness and scalability of Open RE systems in real-world text.

Abstract

Extracting multiple relations from text sentences is still a challenge for current Open Relation Extraction (Open RE) tasks. In this paper, we develop several Open RE models based on the bidirectional LSTM-CRF (BiLSTM-CRF) neural network and different contextualized word embedding methods. We also propose a new tagging scheme to solve overlapping problems and enhance models' performance. From the evaluation results and comparisons between models, we select the best combination of tagging scheme, word embedder, and BiLSTM-CRF network to achieve an Open RE model with a remarkable extracting ability on multiple-relation sentences.

Paper Structure

This paper contains 14 sections, 3 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Tagging Scheme
  5. Model Architecture
  6. Experiments
  7. Data
  8. Word Embeddings
  9. Parameter Settings
  10. Evaluation
  11. Results and Discussion
  12. Results
  13. Discussion
  14. Conclusion

Figures (2)

  • Figure 1: Example of Open Relation Extraction (ARG: Argument, REL: Relation)
  • Figure 2: stanovsky2018supervisedstanovsky2018supervised Tagging Scheme only produces one tag per sentence. For a sentence with multiple relations, there are some overlapping tag representations of one same word across different tagging sequences. In the New Tagging Scheme, relations are annotated orderly so that less O tags are used.