Learning Dynamic Context Augmentation for Global Entity Linking

TL;DR

This paper introduces Dynamic Context Augmentation (DCA), a novel method for collective entity linking that efficiently improves inference by sequentially accumulating context, adaptable with various models and learning strategies.

Contribution

The paper proposes DCA, a simple, effective, one-pass approach for collective entity linking that enhances existing models and reduces computational complexity.

Findings

DCA improves entity linking accuracy across multiple settings.

DCA is effective with supervised and reinforcement learning.

DCA reduces inference time compared to traditional methods.

Abstract

Despite of the recent success of collective entity linking (EL) methods, these "global" inference methods may yield sub-optimal results when the "all-mention coherence" assumption breaks, and often suffer from high computational cost at the inference stage, due to the complex search space. In this paper, we propose a simple yet effective solution, called Dynamic Context Augmentation (DCA), for collective EL, which requires only one pass through the mentions in a document. DCA sequentially accumulates context information to make efficient, collective inference, and can cope with different local EL models as a plug-and-enhance module. We explore both supervised and reinforcement learning strategies for learning the DCA model. Extensive experiments show the effectiveness of our model with different learning settings, base models, decision orders and attention mechanisms.