DataAgent: An Intelligence Agent for Tables Based on Large Language Models

With the development of big data and artificial intelligence, tabular data plays an increasingly important role in fields such as science, finance, and healthcare. Though current spreadsheet systems provide various data processing functions, their complicated operations still pose major difficulties for inexperienced users, especially novices, resulting in frequent mistakes. Therefore, there is an urgent need for a low-learning-curve, easy-to-use tool for tabular data processing. Although recent solutions leveraging large language models (LLMs) have attempted to simplify these tasks, they still lack comprehensive support for table generation and validation. This paper proposes DataAgent, a novel framework based on large language models that integrates natural language task generation with tabular data manipulation. Unlike existing tools, DataAgent systematically addresses key issues of table generation, validation, and manipulation through three core modules: the Schema Project, a dual validation mechanism, and localized data processing. Experimental results demonstrate that DataAgent not only significantly reduces the learning curve for users but also substantially improves the accuracy and reliability of table operations, showcasing its advantages in practical applications.