Learning Dense Representations for Entity Retrieval

TL;DR

This paper introduces a fully learned entity retrieval model using a dual encoder that encodes mentions and entities in the same dense space, enabling fast retrieval without alias tables and outperforming traditional baselines.

Contribution

It presents the first fully learned entity retrieval system based on dual encoders trained solely on Wikipedia anchor links, improving over previous methods.

Findings

Outperforms alias table and BM25 baselines.

Achieves competitive results on TACKBP-2010 dataset.

Retrieves candidates extremely fast and generalizes well.

Abstract

We show that it is feasible to perform entity linking by training a dual encoder (two-tower) model that encodes mentions and entities in the same dense vector space, where candidate entities are retrieved by approximate nearest neighbor search. Unlike prior work, this setup does not rely on an alias table followed by a re-ranker, and is thus the first fully learned entity retrieval model. We show that our dual encoder, trained using only anchor-text links in Wikipedia, outperforms discrete alias table and BM25 baselines, and is competitive with the best comparable results on the standard TACKBP-2010 dataset. In addition, it can retrieve candidates extremely fast, and generalizes well to a new dataset derived from Wikinews. On the modeling side, we demonstrate the dramatic value of an unsupervised negative mining algorithm for this task.