Generating Similarity-Based Recommendations for File Management in Personal Cloud Storage

The challenges associated with file management are numerous and daunting. Users havedifficulty organizing the files they acquire, retrieving files, managing access to shared files, and deleting files that are useless or privacy-sensitive. While work in this field has offered some assistance via tools that help users navigate disorganized file collections and more, they do not address existing and future disorganization in a user's file collection. We seek to close this gap. In this dissertation, we first investigate the organization of file collections in personalcloud storage. We then design new techniques and tools for automatically recommending file- management actions to users in that setting based on notions we develop for characterizing file similarity. We do so in three main investigations. In the first, we examine research participants' perceptions of file pairs in their Google Drive accounts. In the second, we conduct two online user studies asking participants to organize their Google Drive accounts in order to investigate real-time file management and evaluate the tool we developed to assist file management, KondoCloud. In the last, we propose and evaluate a new format for summarizing groups of file management recommendations. Throughout our investigations, we describe how tools offering recommendations of fine-grained file management actions based on similarity can support users in managing their personal information. We conclude by discussing the design implications for future file management tools and identifying how to adapt these tools to commercial settings.