A Systematic Investigation of KB-Text Embedding Alignment at Scale

TL;DR

This paper systematically investigates methods for aligning knowledge base and text embeddings at scale, aiming to enable joint reasoning and improve link prediction, especially for emerging entities like COVID-19.

Contribution

It introduces a novel evaluation framework for KB-text embedding alignment and demonstrates how alignment enhances reasoning and link prediction, especially for new entities.

Findings

Alignment improves link prediction accuracy.

Textual information infusion aids emerging entity reasoning.

Evaluation framework effectively compares alignment methods.

Abstract

Knowledge bases (KBs) and text often contain complementary knowledge: KBs store structured knowledge that can support long range reasoning, while text stores more comprehensive and timely knowledge in an unstructured way. Separately embedding the individual knowledge sources into vector spaces has demonstrated tremendous successes in encoding the respective knowledge, but how to jointly embed and reason with both knowledge sources to fully leverage the complementary information is still largely an open problem. We conduct a large-scale, systematic investigation of aligning KB and text embeddings for joint reasoning. We set up a novel evaluation framework with two evaluation tasks, few-shot link prediction and analogical reasoning, and evaluate an array of KB-text embedding alignment methods. We also demonstrate how such alignment can infuse textual information into KB embeddings for more accurate link prediction on emerging entities and events, using COVID-19 as a case study.