Continual Event Extraction with Semantic Confusion Rectification

TL;DR

This paper introduces a continual event extraction model that addresses semantic confusion and class imbalance by using pseudo labels and knowledge transfer, improving performance on evolving and imbalanced datasets.

Contribution

It proposes a novel model with semantic confusion rectification techniques, including pseudo labels and knowledge transfer, for continual event extraction.

Findings

Outperforms state-of-the-art baselines

Effective on imbalanced datasets

Reduces semantic confusion in event types

Abstract

We study continual event extraction, which aims to extract incessantly emerging event information while avoiding forgetting. We observe that the semantic confusion on event types stems from the annotations of the same text being updated over time. The imbalance between event types even aggravates this issue. This paper proposes a novel continual event extraction model with semantic confusion rectification. We mark pseudo labels for each sentence to alleviate semantic confusion. We transfer pivotal knowledge between current and previous models to enhance the understanding of event types. Moreover, we encourage the model to focus on the semantics of long-tailed event types by leveraging other associated types. Experimental results show that our model outperforms state-of-the-art baselines and is proficient in imbalanced datasets.