DiCoRe: Enhancing Zero-shot Event Detection via Divergent-Convergent LLM Reasoning

2025-06-05

AI Summary

Zero-shot event detection (ZS-ED) faces challenges including complex domain-specific event ontologies, difficulty in trigger identification, and unstable structured output generation; existing LLM-based approaches suffer from limited generalization and accuracy. This paper proposes a decoupled dual-path reasoning framework: a divergent path employs the Dreamer module for open-ended event discovery, while a convergent path integrates Grounder—a finite-state-machine-constrained decoder—with LLM-Judge, a multi-stage verification mechanism, to ensure accurate, domain-adapted structured outputs. Notably, this work is the first to incorporate symbolic constraints into LLM-based zero-shot ED inference. Evaluated across five specialized domains, six benchmark datasets, and nine mainstream LLMs, the framework achieves average F1 improvements of 4–7% over prior state-of-the-art methods, establishing a new baseline for zero-shot event detection.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Zero-shot Event Detection (ED), the task of identifying event mentions in natural language text without any training data, is critical for document understanding in specialized domains. Understanding the complex event ontology, extracting domain-specific triggers from the passage, and structuring them appropriately overloads and limits the utility of Large Language Models (LLMs) for zero-shot ED. To this end, we propose DiCoRe, a divergent-convergent reasoning framework that decouples the task of ED using Dreamer and Grounder. Dreamer encourages divergent reasoning through open-ended event discovery, which helps to boost event coverage. Conversely, Grounder introduces convergent reasoning to align the free-form predictions with the task-specific instructions using finite-state machine guided constrained decoding. Additionally, an LLM-Judge verifies the final outputs to ensure high precision. Through extensive experiments on six datasets across five domains and nine LLMs, we demonstrate how DiCoRe consistently outperforms prior zero-shot, transfer-learning, and reasoning baselines, achieving 4-7% average F1 gains over the best baseline -- establishing DiCoRe as a strong zero-shot ED framework.

Problem

• Identifying event mentions without training data
• Handling complex event ontology and domain triggers
• Improving zero-shot event detection with LLM reasoning

Innovation

• Divergent-convergent reasoning framework DiCoRe
• Dreamer for open-ended event discovery
• Grounder with constrained decoding alignment