Search papers, labs, and topics across Lattice.
This paper introduces CRAFTQA, an adaptive framework designed to enhance reasoning over complex structured data by generating executable Python code sequences tailored to specific questions. By integrating the CodeSTEP module for step-by-step reasoning and the CRAFT module for dynamic function generation, CRAFTQA significantly surpasses existing methods that rely on static predefined functions. Experimental results across various structured datasets reveal substantial improvements in complex reasoning capabilities, underscoring the framework's effectiveness in real-world applications.
CRAFTQA's ability to dynamically generate custom code functions allows it to tackle complex reasoning tasks that traditional methods cannot handle.
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.
