Research map / Graph learning / Knowledge and temporal graphs
A Joint Framework with Heterogeneous-Relation-Aware Graph and Multi-Channel Label Enhancing Strategy for Event Causality Extraction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes JEF-HM, a joint multi-label grid-tagging framework for event causality extraction that combines a heterogeneous-relation-aware graph (R-GAT) and a multi-channel label enhancing module, reaching state-of-the-art F1 on ECE-CCKS.
Problem
Event causality extraction (ECE) must extract cause-effect event pairs with structured arguments, but prior methods ignore explicit event-event causal modeling and use two independent grids for cause and effect events, failing to capture dependencies between the cause/effect extraction subtasks.
Approach
Builds a heterogeneous graph over candidate cause/effect event and argument nodes with event-event, event-argument, and argument-argument edges (R-GAT), and uses a multi-channel biaffine label grid whose preliminary CET and EAE outputs are fed back to enhance interaction between subtasks in a single joint framework.
Key results
- JEF-HM with MacBERT reaches 54.34% ECE F1 on ECE-CCKS, versus 52.66% for DualCor (MacBERT) and 54.08% for fine-tuned ChatGLM-6B.
- JEF-HM[MacBERT] achieves 58.43% EAE F1 and 66.24% CET F1, improvements of 1.47 and 0.79 points over DualCor.
- On the multi-pair subset, JEF-HM improves ECE F1 to 45.14% versus 40.78% for DualCor (with ChatGLM-6B also reported in that subset figure).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the heterogeneous-relation-aware graph module (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 53.10 -> 54.34 | Helped |
| add event-event edges (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 52.72 -> 54.34 | Helped |
| add event-argument edges (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 52.61 -> 54.34 | Helped |
| add argument-argument edges (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 52.62 -> 54.34 | Helped |
| add the multi-channel label enhancing module (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 51.38 -> 54.34 | Helped |
| add the enhancing strategy (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 52.96 -> 54.34 | Helped |
| add auxiliary supervision for multi-channel label representations (ablation) | JEF-HM · ECE-CCKS test | ECE F1: 53.20 -> 54.34 | Helped |
Surprising findings
- Fine-tuned ChatGLM-6B performs best among baselines on EAE and on single-pair samples, yet underperforms on CET and multi-pair samples.
Limitations
- Evaluated only on a single Chinese dataset (ECE-CCKS), with small dataset size cited as limiting graph model gains.
- Compared against only one LLM (ChatGLM-6B); the paper suggests LLMs could further assist ECE.
- Relies on the LTP toolkit for syntactic dependency edges.
Experimental setup
Datasets: ECE-CCKS
Benchmarks: ECE-CCKS
Metrics: Precision Recall Micro-F1
Baselines: Novel-tagging CasECE Pair-linking DualCor ChatGLM-6B
Builds on: DualCor (dual grid tagging) R-GAT MacBERT biaffine attention
Closest papers by meaning
- Logic Induced High-Order Reasoning Network for Event-Event Relation Extraction (AAAI 2025 · similarity 0.87)
- Mastering Context-to-Label Representation Transformation for Event Causality Identification with Diffusion Models (AAAI 2024 · similarity 0.86)
- Dependency Structure-Enhanced Graph Attention Networks for Event Detection (AAAI 2024 · similarity 0.85)
- To Know the Causes of Things: Text Mining for Causal Relations (AAAI 2024 · similarity 0.84)
- Beyond Grounding: Extracting Fine-Grained Event Hierarchies across Modalities (AAAI 2024 · similarity 0.82)
- Cross-modal Multi-task Learning for Multimedia Event Extraction (AAAI 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.