Problem
Research questions and friction points this paper is trying to address.
Spreadsheet Reasoning
Large Language Models
Cross-table Associations
Spatial Layouts
Semi-structured Data
SheetCompass: Hierarchical Relation Graphs for Agentic Spreadsheet Reasoning
🤖 AI Summary
This study addresses the loss of cross-sheet semantics and spatial context caused by structure flattening in large language model-based spreadsheet reasoning. To overcome the limitations of existing serialization methods, this work proposes a graph-guided and memory-driven agent framework. By pioneering a hierarchical relation graph to explicitly model intra- and inter-sheet structures, combined with a memory-augmented mechanism for maintaining global context, the proposed approach effectively preserves multi-dimensional structural semantics. Consequently, it significantly enhances the accuracy and robustness of complex workbook reasoning and automation tasks. Ultimately, this framework establishes a novel paradigm for intelligent spreadsheet processing by mitigating the contextual degradation inherent in conventional linearization techniques.
📝 Abstract
Spreadsheets are widely used to organize, analyze, and manipulate semi-structured data, yet automated spreadsheet reasoning remains challenging for large language models (LLMs). Real-world workbooks often contain implicit cross-table associations, fine-grained column dependencies, and complex spatial layouts. Existing methods typically flatten these multidimensional structures into sequential strings, losing important intra-sheet boundaries and inter-sheet semantics. Consequently, LLMs cannot exploit the global spatial context that human experts naturally use when inspecting spreadsheets. We propose SheetCompass, a graph-guided and memory-driven agentic framework for spreadsheet reasoning and automation. SheetCompass explicitly models structural relationships within and across worksheets while maintaining task-relevant information in memory, enabling agents to reason more effectively over complex workbooks.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Hierarchical Relation Graphs
Agentic Spreadsheet Reasoning
Memory-driven Framework
Cross-table Semantics
Spatial Context
🔎 Similar Papers
No similar papers found.
