Conventional file management systems primarily rely on structured metadata such as filenames, file extensions, and creation dates to manage and organize files. However, such metadata alone fails to capture the actual content or semantic meaning of a file, often leading to results misaligned with user intent. To overcome these limitations, we developed the Content-based File Classification and Organization System (CFCOS), which integrates a Large Language Model (LLM) to perform content-aware file analysis. The LLM generates semantic summaries of file contents and classifies files into meaningful categories based on composition-derived criteria, enabling organization strategies that go beyond rigid, rule-based methods. Through a range of evaluations, we analyze how CFCOS addresses key limitations of conventional file management systems and characterize the properties of LLM-based approaches to content-aware file organization. Furthermore, these results suggest that our approach can be generalized beyond file systems, enabling the semantic and personalized transformation of existing services through prompt engineering.
Son et al. (Mon,) studied this question.