LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention

TL;DR

LUKE introduces a novel transformer-based model that generates contextualized representations for words and entities, improving performance on various entity-related NLP tasks through an entity-aware self-attention mechanism.

Contribution

The paper presents a new pretraining task and an entity-aware self-attention mechanism, leading to state-of-the-art results on multiple entity-centric NLP benchmarks.

Findings

Achieves state-of-the-art on five datasets.

Outperforms previous models in entity typing and relation classification.

Demonstrates effectiveness of entity-aware self-attention.

Abstract

Entity representations are useful in natural language tasks involving entities. In this paper, we propose new pretrained contextualized representations of words and entities based on the bidirectional transformer. The proposed model treats words and entities in a given text as independent tokens, and outputs contextualized representations of them. Our model is trained using a new pretraining task based on the masked language model of BERT. The task involves predicting randomly masked words and entities in a large entity-annotated corpus retrieved from Wikipedia. We also propose an entity-aware self-attention mechanism that is an extension of the self-attention mechanism of the transformer, and considers the types of tokens (words or entities) when computing attention scores. The proposed model achieves impressive empirical performance on a wide range of entity-related tasks. In particular, it obtains state-of-the-art results on five well-known datasets: Open Entity (entity typing), TACRED (relation classification), CoNLL-2003 (named entity recognition), ReCoRD (cloze-style question answering), and SQuAD 1.1 (extractive question answering). Our source code and pretrained representations are available at https://github.com/studio-ousia/luke.