Understanding and Improving Personal File Retrieval
Thesis DisciplineComputer Science
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
Personal file retrieval – the task of locating and opening files on a computer – is a common task for all computer users. A range of interfaces are available to assist users in retrieving files, such as navigation within a file browser, search interfaces and recent items lists. This thesis examines two broad goals in file retrieval: understanding current file retrieval behaviour, and improving file retrieval by designing improved user interfaces.
A thorough understanding of current file retrieval behaviour is important to the design of any improved retrieval tools, however there has been surprisingly little research about the ways in which users interact with common file retrieval tools. To address this, this thesis describes a longitudinal field study that logs participants' file retrieval behaviour across a range of methods, using a specially developed logging tool called FileMonitor. Results confirm findings from previous research that search is used as a method of last resort, while providing new results characterising file retrieval. These include analyses of revisitation behaviour, file browser window reuse, and interactions between retrieval methods, as well as detailed characterisations of the use of navigation and search.
Knowledge gained from this study assists in the design of three improvements to file navigation: Icon Highlights, Search Directed Navigation and Hover Menus. Icon Highlights highlight items that are considered the most likely to be accessed next. These highlights are determined using a new algorithm, AccessRank, which is designed to produce a set of results that is both accurate and stable over time. Search Directed Navigation highlights items that match, or contain items that match, a filename search query, allowing users to rehearse the mechanisms for expert performance in order to aid future retrievals, and providing greater context than the results of a traditional search interface. Hover Menus appear when hovering the mouse cursor above a folder, and provide shortcuts to highly ranked files and folders located at any depth within the folder. This allows users to reduce navigation times by skipping levels of the file hierarchy.
These interfaces are evaluated in lab and field studies, allowing for both precise analysis of their relative strengths and weaknesses, while also providing a high degree of external validity. Results of the lab study show that all three techniques reduce retrieval times and are subjectively preferred by participants. For the field study, fully functional versions of Icon Highlights and Search Directed Navigation are implemented as part of Finder Highlights, a plugin to OS X's file manager. Results indicate that Icon Highlights significantly reduce file retrieval times, and that Search Directed Navigation was useful to those who used it, but faces barriers to adoption.
Key contributions of this thesis include a review of previous literature on file management, a thorough characterisation of file retrieval behaviour, improved algorithms for predicting user behaviour and three improved interfaces for file retrieval. This research has the potential to improve a tedious activity that users perform many times a day, while also providing generalisable algorithms and interface concepts that are applicable to a wide range of interfaces beyond file management.