Title: CRAFTQA: An Adaptive, Code-Driven Framework for Reasoning Over Complex Structured Data

Original: arXiv:2606.02170v1 Announce Type: new Abstract: Real-world scenarios involve massive heterogeneous structured data (e.g., tables, knowledge graphs), making effective reasoning over such diverse data increasingly important. Unified structured data question answering has emerged as a prominent research trend, aiming to answer natural language questions across different structured data types within a single framework. However, existing unified methods share a common limitation: they rely on a set of predefined functions, which restricts their ability to perform complex reasoning beyond these predefined operations. To overcome this fundamental limitation, we propose CRAFTQA, a novel adaptive code-driven framework comprising two core modules, CodeSTEP and CRAFT. The CodeSTEP module is a paradigm that generates a complete executable Python code sequence, which contains step-by-step code-based reasoning operations based on the question. The CRAFT module dynamically generates custom code functions for operations beyond the predefined function set, and seamlessly integrates with CodeSTEP to significantly enhance flexibility in handling complex reasoning. Comprehensive experiments on multiple structured datasets demonstrate that CRAFTQA achieves remarkable improvements in complex reasoning scenarios compared to existing unified methods.

Rewrite: As real-world applications increasingly generate vast amounts of heterogeneous structured data—such as knowledge graphs and tables—the capacity to reason effectively over these diverse formats has become paramount. Consequently, unified structured data question answering has gained traction as a key research area, striving to address natural language queries across various data structures within a cohesive system. Despite this progress, current unified approaches suffer from a significant constraint: they depend on fixed sets of predefined functions, thereby limiting their ability to execute complex reasoning tasks that fall outside these predefined boundaries.

To address this critical shortcoming, we introduce CRAFTQA, an innovative adaptive framework driven by code generation. This system is built upon two primary components: CodeSTEP and CRAFT. The CodeSTEP component operates as a paradigm that produces a fully executable Python script, outlining step-by-step reasoning operations derived directly from the user's query. Complementing this, the CRAFT module dynamically creates custom code functions for any operations not covered by the standard predefined set. By integrating seamlessly with CodeSTEP, CRAFT substantially boosts the framework's adaptability and flexibility when tackling intricate reasoning challenges. Extensive evaluations across multiple structured datasets confirm that CRAFTQA delivers substantial performance gains in complex reasoning contexts when compared to prevailing unified methods.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC