Research map / Graph learning / Knowledge and temporal graphs
Logic Induced High-Order Reasoning Network for Event-Event Relation 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 LogicERE, which builds a logic constraint induced graph (event and event-pair nodes, no external tools) with a relational graph transformer and adds symmetry and conjunction logic losses to jointly extract temporal and subevent event relations.
Problem
Event-event relation extraction (temporal and subevent relations) in documents, where existing event graphs rely on heuristic external tools and fail to preserve logical constraints (coreference, symmetry, conjunction) among relations.
Approach
Builds a graph with both event nodes and event-pair nodes, whose edges encode coreference, symmetry and conjunction constraints; reasons over it with a relational graph transformer and enforces logic constraints via differentiable product t-norm losses in joint training.
Key results
- Reaches 83.7 F1 on MATRES (vs. 83.1 for SDLG) and 88.7 F1 on TCR (vs. 87.6 for SDLG).
- Achieves 65.4 average F1 on HiEve (vs. 62.0 for OntoEnhance).
- On MAVEN-ERE joint setting, 55.0 F1 for TRE and 30.3 F1 for SRE, improving SRE by 3.0 points over the best baseline.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add edge heterogeneity to the relational aggregation (ablation) | LogicERE · MAVEN-ERE TRE | F1: 54.4 -> 55.0 | Helped |
| Add edge heterogeneity to the relational aggregation (ablation) | LogicERE · MAVEN-ERE SRE | F1: 29.5 -> 30.3 | Helped |
| Add coreference edges and use coreference training labels (ablation) | LogicERE · MAVEN-ERE TRE | F1: 54.7 -> 55.0 | Helped |
| Add coreference edges and use coreference training labels (ablation) | LogicERE · MAVEN-ERE SRE | F1: 30.0 -> 30.3 | Helped |
| Add event-event pair edges (ablation) | LogicERE · MAVEN-ERE TRE | F1: 53.5 -> 55.0 | Helped |
| Add event-event pair edges (ablation) | LogicERE · MAVEN-ERE SRE | F1: 28.8 -> 30.3 | Helped |
| Add the symmetry objective (ablation) | LogicERE · MAVEN-ERE TRE | F1: 54.8 -> 55.0 | Helped |
| Add the symmetry objective (ablation) | LogicERE · MAVEN-ERE SRE | F1: 30.0 -> 30.3 | Helped |
| Add the conjunction objective (ablation) | LogicERE · MAVEN-ERE TRE | F1: 54.5 -> 55.0 | Helped |
| Add the conjunction objective (ablation) | LogicERE · MAVEN-ERE SRE | F1: 29.7 -> 30.3 | Helped |
| Add joint logic learning (ablation) | LogicERE · MAVEN-ERE TRE | F1: 54.1 -> 55.0 | Helped |
| Add joint logic learning (ablation) | LogicERE · MAVEN-ERE SRE | F1: 28.9 -> 30.3 | Helped |
Limitations
- Relies on a pre-trained coreference model to derive coreference edges for MATRES, which lacks gold coreference annotations.
- SRE performance remains low on MAVEN-ERE (around 30 F1).
Experimental setup
Datasets: MATRES TCR HiEve MAVEN-ERE
Benchmarks: MATRES TCR HiEve MAVEN-ERE
Metrics: precision recall F1
Baselines: TEMPROB+ILP Hierarchical Joint Constrain Self-Training Vanilla Classifier Relative Time Probabilistic Box TGAGCN OntoEnhance SDLG StructLR TACOLM
Builds on: relational graph transformer logic-driven framework for consistency (constrained learning) RoBERTa
Closest papers by meaning
- A Joint Framework with Heterogeneous-Relation-Aware Graph and Multi-Channel Label Enhancing Strategy for Event Causality Extraction (AAAI 2024 · similarity 0.87)
- TEILP: Time Prediction over Knowledge Graphs via Logical Reasoning (AAAI 2024 · similarity 0.85)
- A Comprehensive Evaluation on Event Reasoning of Large Language Models (AAAI 2025 · similarity 0.85)
- Text2NKG: Fine-Grained N-ary Relation Extraction for N-ary relational Knowledge Graph Construction (NeurIPS 2024 · similarity 0.84)
- Reasoning of Large Language Models over Knowledge Graphs with Super-Relations (ICLR 2025 · similarity 0.84)
- Latent Logic Tree Extraction for Event Sequence Explanation from LLMs (ICML 2024 · similarity 0.84)
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.