Traversing Knowledge Graphs in Vector Space

TLDR

Path queries on knowledge graphs can answer compositional questions, but missing facts disrupt them; recent embedding models impute missing edges by embedding the graph in vector space. The authors aim to develop a compositional training objective that enhances models’ ability to answer path queries. They apply this compositional objective to existing embedding models, training them to recursively compose relations in vector space. The new objective more than doubles path‑query accuracy, reduces overall errors by up to 43%, and yields state‑of‑the‑art performance on standard knowledge‑base completion tasks, though recursive application still incurs cascading errors.

Abstract

Path queries on a knowledge graph can be used to answer compositional questions such as "What languages are spoken by people living in Lisbon?" However, knowledge graphs often have missing facts (edges) which disrupts path queries. Recent models for knowledge base completion impute missing facts by embedding knowledge graphs in vector spaces. We show that these models can be recursively applied to answer path queries, but that they suffer from cascading errors. This motivates a new "compositional" training objective, which dramatically improves all models' ability to answer path queries, in some cases more than doubling accuracy. On a standard knowledge base completion task, we also demonstrate that compositional training acts as a novel form of structural regularization, reliably improving performance across all base models (reducing errors by up to 43%) and achieving new state-of-the-art results.