Can KBQA Models Predict Their Reasoning Paths? Isomorphism Prediction Task as a Proxy

Zhen Wu; Ritam Dutt; Dhruv Gupta; Carolyn Rose

Can KBQA Models Predict Their Reasoning Paths? Isomorphism Prediction Task as a Proxy

Zhen Wu, Ritam Dutt, Dhruv Gupta, Carolyn Rose

Published: 06 Apr 2025, Last Modified: 18 Apr 2025LTI-SRS 2025 PosterEveryoneRevisionsBibTeXCC BY 4.0

Track: Main Track

Keywords: knowledge base QA, reasoning, generalization

TL;DR: We introduce isomorphism prediction to enhance reasoning fidelity in KBQA and show that it improves generalization through contrastive co-distillation and benefits KBQA performance in a multitask setting.

Abstract: Despite achieving correct answers, we find that existing Knowledge Base Question Answering (KBQA) models struggle to follow the expected reasoning structures. We introduce the task of isomorphism prediction to enhance reasoning fidelity beyond answer generation, with a focus on generalization. We propose a contrastive knowledge co-distillation framework that unifies textual and graphical KBQA paradigms, improving isomorphism prediction and overall model generalization. Furthermore, incorporating isomorphism prediction as an auxiliary task could also improve KBQA performance.

Submission Number: 9

Loading