Beyond Seen Data: Improving KBQA Generalization Through Schema-Guided Logical Form Generation

TL;DR

This paper presents SG-KBQA, a schema-guided model that improves the generalization of KBQA systems to unseen knowledge base elements by incorporating schema contexts into logical form generation.

Contribution

It introduces a novel schema-guided approach for entity retrieval and logical form generation to enhance KBQA generalization to unseen data.

Findings

Outperforms state-of-the-art models on benchmark datasets.

Achieves strong generalizability across diverse test settings.

Effectively utilizes schema contexts to handle unseen KB elements.

Abstract

Knowledge base question answering (KBQA) aims to answer user questions in natural language using rich human knowledge stored in large KBs. As current KBQA methods struggle with unseen knowledge base elements at test time,we introduce SG-KBQA: a novel model that injects schema contexts into entity retrieval and logical form generation to tackle this issue. It uses the richer semantics and awareness of the knowledge base structure provided by schema contexts to enhance generalizability. We show that SG-KBQA achieves strong generalizability, outperforming state-of-the-art models on two commonly used benchmark datasets across a variety of test settings. Our source code is available at https://github.com/gaosx2000/SG_KBQA.