KC-GEE: Knowledge-based Conditioning for Generative Event Extraction

Abstract

Event extraction is an important but challenging task. Many existing techniques decompose it into event and argument detection/classification subtasks, which are complex structured prediction problems. Generation-based extraction techniques lessen the complexity of the problem formulation and are able to leverage the reasoning capabilities of large pretrained language models. However, they still suffer from poor zero-shot generalizability and are ineffective in handling long contexts such as documents. We propose a generative event extraction model, TC-GEE, that addresses these limitations. A key contribution of TC-GEE is a novel knowledge-based conditioning technique that injects the schema of candidate event types as the prefix into each layer of an encoder-decoder language model, thus enabling effective zero-shot learning and improving supervised learning. Our experiments on two benchmark datasets demonstrate the strong performance of our TC-GEE model. It achieves especially strong performance in the challenging document-level extraction task and in the zero-shot learning setting, outperforming state-of-the-art models by up to 27.7 absolute F1 points.

Theia: Weakly Supervised Multimodal Event Extraction from Incomplete Data

“…With the advances in deep learning, models based on neural networks have also been developed for this task (Nguyen and Nguyen, 2019;Zhang et al, 2019). More recently, several studies leverage the strong representation and reasoning capabilities of pre-trained language models (Li et al, 2021; Wu et al, 2022;Hsu et al, 2022).…”