Efficient One-Pass End-to-End Entity Linking for Questions

TL;DR

ELQ is a fast, end-to-end entity linking model for questions that jointly detects and links entities in one pass, significantly improving accuracy and inference speed for question answering systems.

Contribution

The paper introduces ELQ, a novel one-pass biencoder model that jointly performs mention detection and linking for questions, achieving state-of-the-art results and fast inference.

Findings

ELQ outperforms previous methods by +12.7% and +19.6% F1 on WebQSP and GraphQuestions.

ELQ achieves a very fast inference speed of 1.57 examples/sec on a CPU.

Using ELQ improves downstream question answering performance.

Abstract

We present ELQ, a fast end-to-end entity linking model for questions, which uses a biencoder to jointly perform mention detection and linking in one pass. Evaluated on WebQSP and GraphQuestions with extended annotations that cover multiple entities per question, ELQ outperforms the previous state of the art by a large margin of +12.7% and +19.6% F1, respectively. With a very fast inference time (1.57 examples/s on a single CPU), ELQ can be useful for downstream question answering systems. In a proof-of-concept experiment, we demonstrate that using ELQ significantly improves the downstream QA performance of GraphRetriever (arXiv:1911.03868). Code and data available at https://github.com/facebookresearch/BLINK/tree/master/elq