KEPLER: A Unified Model for Knowledge Embedding and Pre-trained Language Representation

TL;DR

KEPLER is a unified model that integrates knowledge embedding with pre-trained language models, enhancing factual knowledge representation and text-based KE, achieving state-of-the-art results in NLP tasks and KG link prediction.

Contribution

This paper introduces KEPLER, a novel unified model combining KE and PLMs, and provides a large-scale KG dataset Wikidata5M for benchmarking and research.

Findings

Achieves state-of-the-art performance on various NLP tasks.

Works effectively as an inductive KE model for KG link prediction.

Provides a new large-scale KG benchmark dataset, Wikidata5M.

Abstract

Pre-trained language representation models (PLMs) cannot well capture factual knowledge from text. In contrast, knowledge embedding (KE) methods can effectively represent the relational facts in knowledge graphs (KGs) with informative entity embeddings, but conventional KE models cannot take full advantage of the abundant textual information. In this paper, we propose a unified model for Knowledge Embedding and Pre-trained LanguagE Representation (KEPLER), which can not only better integrate factual knowledge into PLMs but also produce effective text-enhanced KE with the strong PLMs. In KEPLER, we encode textual entity descriptions with a PLM as their embeddings, and then jointly optimize the KE and language modeling objectives. Experimental results show that KEPLER achieves state-of-the-art performances on various NLP tasks, and also works remarkably well as an inductive KE model on KG link prediction. Furthermore, for pre-training and evaluating KEPLER, we construct Wikidata5M, a large-scale KG dataset with aligned entity descriptions, and benchmark state-of-the-art KE methods on it. It shall serve as a new KE benchmark and facilitate the research on large KG, inductive KE, and KG with text. The source code can be obtained from https://github.com/THU-KEG/KEPLER.