Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Complex Causality Extraction via Improved Subtask Interaction and Knowledge Fusion

Jinglong Gao, Chen Lu, Xiao Ding, Zhongyang Li, Ting Liu, Bing Qin

TL;DR

This work targets Event Causality Extraction (ECE) by addressing three core challenges: Complex Causality Extraction with multiple cause–effect pairs per sentence, Subtask Interaction between Event Extraction (EE) and Event Causality Identification (ECI), and Knowledge Fusion between pretrained language models (PLMs) and knowledge graphs (KGs). It introduces UniCE, a unified, multi-layer framework with an Event Module and a Relation Module that interact across layers, aided by a background graph constructed from sentence-linked KG nodes and dynamic insertion of extracted events. Key innovations include the insertion induction module (Kirchhoff Matrix-Tree Theorem-based graph connections), the T-aggregator and K-aggregator for cross-subtask and cross-modality fusion, and separate decoders for EE and ECI to handle complex causality. Experiments on ESC, SCIFI, and CTB show state-of-the-art performance and and a substantial margin over ChatGPT, with additional evidence that UniCE can boost ChatGPT performance via in-context learning; however, the approach incurs slower inference due to KG reasoning, suggesting a scheduling-based mitigation for speed-sensitive applications.

Abstract

Event Causality Extraction (ECE) aims at extracting causal event pairs from texts. Despite ChatGPT's recent success, fine-tuning small models remains the best approach for the ECE task. However, existing fine-tuning based ECE methods cannot address all three key challenges in ECE simultaneously: 1) Complex Causality Extraction, where multiple causal-effect pairs occur within a single sentence; 2) Subtask~ Interaction, which involves modeling the mutual dependence between the two subtasks of ECE, i.e., extracting events and identifying the causal relationship between extracted events; and 3) Knowledge Fusion, which requires effectively fusing the knowledge in two modalities, i.e., the expressive pretrained language models and the structured knowledge graphs. In this paper, we propose a unified ECE framework (UniCE to address all three issues in ECE simultaneously. Specifically, we design a subtask interaction mechanism to enable mutual interaction between the two ECE subtasks. Besides, we design a knowledge fusion mechanism to fuse knowledge in the two modalities. Furthermore, we employ separate decoders for each subtask to facilitate complex causality extraction. Experiments on three benchmark datasets demonstrate that our method achieves state-of-the-art performance and outperforms ChatGPT with a margin of at least 30% F1-score. More importantly, our model can also be used to effectively improve the ECE performance of ChatGPT via in-context learning.

Enhancing Complex Causality Extraction via Improved Subtask Interaction and Knowledge Fusion

TL;DR

This work targets Event Causality Extraction (ECE) by addressing three core challenges: Complex Causality Extraction with multiple cause–effect pairs per sentence, Subtask Interaction between Event Extraction (EE) and Event Causality Identification (ECI), and Knowledge Fusion between pretrained language models (PLMs) and knowledge graphs (KGs). It introduces UniCE, a unified, multi-layer framework with an Event Module and a Relation Module that interact across layers, aided by a background graph constructed from sentence-linked KG nodes and dynamic insertion of extracted events. Key innovations include the insertion induction module (Kirchhoff Matrix-Tree Theorem-based graph connections), the T-aggregator and K-aggregator for cross-subtask and cross-modality fusion, and separate decoders for EE and ECI to handle complex causality. Experiments on ESC, SCIFI, and CTB show state-of-the-art performance and and a substantial margin over ChatGPT, with additional evidence that UniCE can boost ChatGPT performance via in-context learning; however, the approach incurs slower inference due to KG reasoning, suggesting a scheduling-based mitigation for speed-sensitive applications.

Abstract

Event Causality Extraction (ECE) aims at extracting causal event pairs from texts. Despite ChatGPT's recent success, fine-tuning small models remains the best approach for the ECE task. However, existing fine-tuning based ECE methods cannot address all three key challenges in ECE simultaneously: 1) Complex Causality Extraction, where multiple causal-effect pairs occur within a single sentence; 2) Subtask~ Interaction, which involves modeling the mutual dependence between the two subtasks of ECE, i.e., extracting events and identifying the causal relationship between extracted events; and 3) Knowledge Fusion, which requires effectively fusing the knowledge in two modalities, i.e., the expressive pretrained language models and the structured knowledge graphs. In this paper, we propose a unified ECE framework (UniCE to address all three issues in ECE simultaneously. Specifically, we design a subtask interaction mechanism to enable mutual interaction between the two ECE subtasks. Besides, we design a knowledge fusion mechanism to fuse knowledge in the two modalities. Furthermore, we employ separate decoders for each subtask to facilitate complex causality extraction. Experiments on three benchmark datasets demonstrate that our method achieves state-of-the-art performance and outperforms ChatGPT with a margin of at least 30% F1-score. More importantly, our model can also be used to effectively improve the ECE performance of ChatGPT via in-context learning.
Paper Structure (30 sections, 7 equations, 4 figures, 6 tables)

This paper contains 30 sections, 7 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview of UniCE
  4. Event Module
  5. Relation Module
  6. Background Graph Construction:
  7. Reasoning over Graph
  8. Causal Relation Classifier
  9. Insertion Induction
  10. Subtask Information Aggregator
  11. Knowledge Information Aggregator
  12. Multi-layer Learning & Inference
  13. Experiments
  14. Experimental Setup
  15. Dataset and Evaluation Metrics
  16. ...and 15 more sections

Figures (4)

  • Figure 1: An example of the ECE task. All the arrows are from the cause event to the effect event.
  • Figure 2: The illustration of the proposed unified ECE framework UniCE.
  • Figure 3: Experimental results on the ECE task with varying numbers of causal pairs per sentence.
  • Figure 4: Case study of examples from SCIFI and EventStoryLine datasets.