Pandora: Leveraging Code-Driven Knowledge Transfer for Unified Structured Knowledge Reasoning

Yongrui Chen

doi:10.1109/tkde.2026.3718049

Pandora: Leveraging Code-Driven Knowledge Transfer for Unified Structured Knowledge Reasoning

2026-07-29

0

PRE

AI

DOI：10.1109/tkde.2026.3718049

Original

Original request for help

Save

Abstract

En 中文

Unified Structured Knowledge Reasoning (USKR) aims to answer natural language questions by using structured sources such as tables, databases, and knowledge graphs in a unified way. Existing USKR methods rely on task-specific strategies or bespoke representations, which hinder their ability to dismantle barriers between different SKR tasks, thereby constraining their overall performance in cross-task scenarios. In this paper, we introduce Pandora, a novel USKR framework that addresses the limitations of existing methods by leveraging two key innovations. First, we propose a code-based unified knowledge representation using Python’s Pandas API, which aligns seamlessly with the pre-training of LLMs. This representation facilitates a cohesive approach to handling different structured knowledge sources. Building on this foundation, we employ knowledge transfer to bolster the unified reasoning process of LLMs by automatically building cross-task memory. By leveraging multi-stage code-driven reasoning and adaptively refining its outputs through execution feedback, Pandora demonstrates strong and unified reasoning capabilities across heterogeneous data sources. Extensive experiments on seven widely used benchmarks across Text-to-SQL, KGQA, and TableQA demonstrate that Pandora outperforms existing unified reasoning frameworks and competes effectively with task-specific methods.

Keywords:

Structured knowledge reasoning

large language model

unified knowledge representation

database interface

knowledge graph question answering

Pandora: Leveraging Code-Driven Knowledge Transfer for Unified Structured Knowledge Reasoning

Abstract

Journal

Organization