http://doc.trolltech.com/pictures.html
Michael MCGuffin
March 18, 2001
The author proposes the implementation of a file browser offering a dual, split view of files, composed of (i) an expandable tree list, showing the hierarchical arrangement of folders, and (ii) a tabletop view of the currently selected folder, with files arranged on its 2D surface. Users will be free to rearrange files on the tabletop, as well as pan and zoom over the 2D surface. The spatial layout of files in each folder will be saved to persist across sessions. This project will explore the potential advantages of enriching the representation of a file system's structure with 2D spatial information.
Digital documents can be automatically sorted, rearranged, categorized, scanned for keywords, and presented to the user in many different forms (e.g. as a list, possibly ordered, of files of a given type or matching a given query; as icons, arranged in rows and columns or freely arrangeable; as links on a webpage; etc.) The ability to automatically reorganize documents and invoke different views and filters is one of the principle virtues of digital files over their physical counterparts.
Nevertheless, the underlying implementation of a file system is almost always hierarchical. Hierarchies are a rigid way of organizing documents, in that they forbid files from being placed in more than one folder. Furthermore, in light of the many different views that a user may create of their files, the underlying organization of the file system may seem somewhat irrelevant. However, this underlying organization serves an important role in giving definiteness to the user's notion of where files are located. Barreau and Nardi [3] observed that, when searching for documents in their personal workspace, users "overwhelmingly" prefer location-based search (i.e. taking a guess at what folder might contain the file and browsing until they find it) over logical search (i.e. invoking an automated search by keyword). Barreau and Nardi also found that, overall, users give names to files not "for the purpose of searching for a particular name" but rather to jog their memory "when scanning files at a location".
Hierarchies give location to files by relating them to each other (e.g. with respect to siblings and parents), engaging contextual memory in users. Hierarchies do not, however, engage spatial memory. Czerwinski et al. [5] have demonstrated the effectiveness of leveraging spatial memory for managing collections of digital documents. The use of spatial relationships also helps users to organize information. "Users like to group items spatially", "have difficulty deciding where to file a new item", and prefer not to immediately categorize new information "into specific folders" [8].
The ability to freely arrange items on a 2D surface enables users to group items while leaving the "structure implicit and informal" [9]. Users can exploit subtle visual cues and experiment with different organizations, allowing them to "delay segmentation [...] of information" and the "organizational commitments" [7] involved in creating folders. Indeed, during Mander's study [8], "Users were sometimes dissatisfied with using folders [...] because they were required to make an explicit decision about how to categorize individual terms".
Spatial arrangement allows users to enrich the file system with semantic information not expressible with folder structure. Barreau and Nardi [3] give anecdotal examples of users arranging icons on their desktop to express urgency of tasks, or as reminders, e.g. "moving icons near the trash can as a reminder to delete them".
Of course, virtual desktops are of limited size, and can quickly become filled with icons. This is a problem particularly for what Barreau and Nardi [3] call "ephemeral information", i.e. files that are needed for only a short time and are not worth the effort of filing.
I propose creating a file browser that displays every folder as a distinct 2D space, with its contents laid out spatially. Users will be free to rearrange the files in each folder, and the spatial arrangement given will persist across sessions. Furthermore, the 2D spatial view will be complemented by a simultaneous expandable tree view of the overall folder hierarchy.
Note that Mac OS already supports the 2D spatial arrangement of icons within every folder. When a folder is opened for browsing, a separate window is popped up to display its contents. Each folder has a corresponding window size and position that is persistent across sessions. When browsing many folders, windows quickly proliferate and have to be closed or moved aside to browse sibling or parent folders. Mac OS also supports an expandable tree list view, for browsing the folder hierarchy structure. The tree list view does not display the spatial layout of files within folders.
The browser I propose differs from Mac OS's interface in that the tree list view and 2D spatial layout are displayed simultaneously. Selecting a folder in the tree list will cause the 2D viewer to display the contents of that folder. This will allow fast browsing of the spatial layout of files with no extra windows being opened. Furthermore, I propose allowing the user to pan and zoom over the 2D view of the folder, effectively increasing the space available to the user and making it easy for spatial arrangements to grow in any direction.
A browser very similar to the one proposed here was evaluated by Amento et al. [1]. Its purpose was to help users in the task of organizing web sites. The browser proposed here differs in that it is intended for a general file system, and will allow users to pan and zoom over the 2D spatial layout of files.
To ease implementation, files in the 2D space will be represented by text strings. Meaningful icons would be difficult to generate automatically, and the study by Czerwinski et al. [5] suggests that icons are not as important as spatial arrangement for leveraging users' spatial memory.
One unobvious choice to make is how files should be arranged by default -- in particular, new files that are created in a folder whose contents have been pre-arranged by the user. One possibility is to always display files that haven't been explicitly positioned by the user in a grid below the files that have been. It may also prove useful to colour-code files that have not been given an explicit position, to flag the user that they are new and have not been spatially organized.
It remains to be seen if the proposed browser would help users manage "large quantities of ephemeral information" [3]. Since each folder will contain a large, distinct 2D space, users will be able to work outside the limited space of their desktop without having to file away all items in a formal, explicit way. However, large numbers of ephemeral files may still pose problems unsolved by the proposed browser.
Also of interest is to explore other forms of spatial layout, such as 3D space [2,4,11]. A more challenging question to explore is whether non-hierarchical schemes can be used to organize files while still supporting browsing and conveying a sense of location to users [6,8,10].
Finally, it would be valuable to enrich the spatial representation of the files so their form would reveal more about their content. Physical books, for example, immediately convey how much information they are likely to contain from their very size.
The file browser will be implemented using Qt, a cross-platform C++ GUI framework available under the GPL (GNU Public License). Qt has support for many sophisticated aggregate widgets, including browsers for simple HTML documents and expandable tree lists.
http://doc.trolltech.com/pictures.html
The implementation will be done on Linux.
The spatial positions of files in each folder will be stored in a hidden "dot file" within the appropriate folder. Thus, if the user moves a subdirectory using a shell command, the spatial data will be transparently carried with it. This also eliminates the need for a central database and the synchronization issues that would arise from one.