Towards Event Extraction with Massive Types: LLM-based Collaborative Annotation and Partitioning Extraction

TL;DR

This work tackles event extraction with massive type schemas by introducing an LLM-based collaborative annotation workflow that refines distant supervision signals, improves trigger/type/argument annotations, and produces the large-scale EEMT dataset (over 200k samples, 3,465 event types, 6,297 roles). It further presents LLM-PEE, a partitioning-based extraction framework that recalls candidate types, partitions them to fit limited context lengths, and extracts events via LLMs, achieving strong supervised gains (ED up to 5.4% F1 and EAE up to 6.1% F1) and notable zero-shot improvements (up to 12.9% F1) over mainstream LLMs. The approach demonstrates improved annotation quality (per-step F1 > 85%, with refined event types reaching 96.2%) and superior generalization, suggesting promising directions for scalable EE across domains. Together, the collaborative annotation and partitioning extraction enable effective handling of massive event type schemas and provide a resource and framework for future EE research and applications.

Abstract

Developing a general-purpose extraction system that can extract events with massive types is a long-standing target in Event Extraction (EE). In doing so, the challenge comes from two aspects: 1) The absence of an efficient and effective annotation method. 2) The absence of a powerful extraction method can handle massive types. For the first challenge, we propose a collaborative annotation method based on Large Language Models (LLMs). Through collaboration among multiple LLMs, it first refines annotations of trigger words from distant supervision and then carries out argument annotation. Next, a voting phase consolidates the annotation preferences across different LLMs. Finally, we create the EEMT dataset, the largest EE dataset to date, featuring over 200,000 samples, 3,465 event types, and 6,297 role types. For the second challenge, we propose an LLM-based Partitioning EE method called LLM-PEE. To overcome the limited context length of LLMs, LLM-PEE first recalls candidate event types and then splits them into multiple partitions for LLMs to extract events. The results in the supervised setting show that LLM-PEE outperforms the state-of-the-art methods by 5.4 in event detection and 6.1 in argument extraction. In the zero-shot setting, LLM-PEE achieves up to 12.9 improvement compared to mainstream LLMs, demonstrating its strong generalization capabilities.

Figures (6)

• Figure 1: Statistics on the existing EE datasets.
• Figure 2: Overview of the proposed LLM-based collaborative annotation method for EE with massive types.
• Figure 3: Overview of our LLM-based Partitioning Extraction Framework, including Event Detection and Event Argument Extraction.
• Figure 4: The argument distribution in our dataset.
• Figure 5: An example of training data in ED task.