SheetCopilot: Bringing Software Productivity to the Next Level through Large Language Models

TL;DR

SheetCopilot leverages large language models to automate spreadsheet tasks through a novel agent that interprets natural language commands, significantly improving task completion accuracy over baseline methods.

Contribution

Introduces a SheetCopilot agent with a set of atomic actions and a state machine framework for robust spreadsheet control using LLMs, along with a curated dataset and evaluation pipeline.

Findings

Achieves 44.3% task completion rate with a single LLM generation.

Outperforms baseline code generation methods.

Provides a dataset of 221 spreadsheet control tasks.

Abstract

Computer end users have spent billions of hours completing daily tasks like tabular data processing and project timeline scheduling. Most of these tasks are repetitive and error-prone, yet most end users lack the skill to automate these burdensome works. With the advent of large language models (LLMs), directing software with natural language user requests become a reachable goal. In this work, we propose a SheetCopilot agent that takes natural language task and control spreadsheet to fulfill the requirements. We propose a set of atomic actions as an abstraction of spreadsheet software functionalities. We further design a state machine-based task planning framework for LLMs to robustly interact with spreadsheets. We curate a representative dataset containing 221 spreadsheet control tasks and establish a fully automated evaluation pipeline for rigorously benchmarking the ability of LLMs in software control tasks. Our SheetCopilot correctly completes 44.3\% of tasks for a single generation, outperforming the strong code generation baseline by a wide margin. Our project page:https://sheetcopilot.github.io/.