Evaluating Design Decisions for Dual Encoder-based Entity Disambiguation

TL;DR

This paper evaluates key design choices in Dual Encoder-based entity disambiguation, introducing VerbalizED, a model that incorporates contextual label verbalizations and hard negative sampling, achieving state-of-the-art results on ZELDA.

Contribution

It systematically assesses design decisions for Dual Encoders in entity disambiguation and proposes VerbalizED with novel features that improve performance.

Findings

VerbalizED achieves new state-of-the-art on ZELDA.

Contextual label verbalizations enhance disambiguation accuracy.

Hard negative sampling improves model robustness.

Abstract

Entity disambiguation (ED) is the task of linking mentions in text to corresponding entries in a knowledge base. Dual Encoders address this by embedding mentions and label candidates in a shared embedding space and applying a similarity metric to predict the correct label. In this work, we focus on evaluating key design decisions for Dual Encoder-based ED, such as its loss function, similarity metric, label verbalization format, and negative sampling strategy. We present the resulting model VerbalizED, a document-level Dual Encoder model that includes contextual label verbalizations and efficient hard negative sampling. Additionally, we explore an iterative prediction variant that aims to improve the disambiguation of challenging data points. Comprehensive experiments on AIDA-Yago validate the effectiveness of our approach, offering insights into impactful design choices that result in a new State-of-the-Art system on the ZELDA benchmark.