Enabling Collaborative Tabletop File System Interaction Through Similarity-based Browsing

Anthony Collins
University of Sydney, Australia
anthony@it.usyd.edu.au

1. Problem and Motivation

Accessing personal digital artefacts, such as files and e-mail, is a common task that modern computer users perform every day for a variety of purposes, and they often need to share this digital information with others in a collocated context. However, with only a single keyboard and mouse present, current computer systems have limited support for multi-user interaction. The interactive tabletop, a novel medium that has recently attracted significant research interest, has the potential to support people collaborating with their files while sitting face-to-face and interacting concurrently. While people may not perceive a tabletop interface as a conventional computer, a form of file system interface is important: a file system interface constitutes a core facility of an operating system. Given the context of use of tabletops — which may often be in social, open environments for collaborating — we need to support multi-user file system access. However, this poses a significant challenge, as hierarchical file systems were originally designed for single-user interaction with a keyboard and mouse [14]. In this research, we reconsider the original metaphors and techniques of file system interaction, to explore the issues in designing effective file system interfaces for tabletop collaboration.

When collaborating with personal files using a shared workspace, people need to access their file systems to locate information relevant to the discussion, even though other users may be also accessing differently organised file systems. While a user may have a reasonable knowledge of the organisation of their own file system, it is potentially difficult to share parts of the file system with others who are unfamiliar with its hierarchical organisation. This poses a key design challenge in supporting people sharing their files at a tabletop. Assisted retrieval of all related information across multiple people's file systems is an important facility to be explored in tabletop file system interfaces. For example, consider two work colleagues who have met to discuss an important project they are working on. They would like to share all project information relevant to the discussion with each other (stored on their personal computers) at an interactive tabletop. The tabletop should be able to support this collaboration scenario by making it easy to access all relevant information regardless of which computer system it is stored on, and where in the file system it is located.

The hierarchical file system has been standard in conventional computer operating system since the 1970s [14], and users have a deep existing mental model of hierarchical file system interaction. Yet, hierarchies are often inadequate for organising the diverse collections of personal information that people need to store [6], and there is no evidence to suggest that hierarchies are the best approach for storing this information on personal computing devices. Hierarchies impose a strict single-classification scheme on information, some of which may be difficult to categorise, or may need to be retrieved from different perspectives [7]. Related information may be stored in multiple folders or file systems, and the retrieval of this requires knowledge of exactly where the information is stored in each file system. While there are facilities to simplify finding 'similar' information in popular web search engines (such as the Google "similar pages" feature [11]), this kind of associative navigation has not been explored for file system interaction.

The problems of hierarchical file system interaction become more acute on novel interaction hardware — despite the advantages of the tabletop medium for collocated collaboration, the constraints call for re-thinking standard approaches to accessing files. First, it must be natural to interact with the tabletop using input with special constraints not present in personal computer interaction: a keyboard and mouse is typically not present in a multi-user tabletop setting. While providing a projected keyboard on the tabletop surface is possible, this provides no tactile feedback when pressing keys, and does not give users a fixed reference of where to place their hands. Interface elements must also be orientation-independent to support people sitting both face-to-face and around-the-table. This makes the use of text (such as filenames) on a tabletop problematic. User interface selection targets must be sufficiently large for direct-touch interaction, which potentially contributes to clutter on the tabletop. Furthermore, enabling tabletop access to increasingly large personal file systems is a key challenge when designing to minimise clutter.

In this paper, we present a novel design for a new way to access files that helps people share similar information with each other, and that operates effectively within the constraints of the tabletop medium. We then outline results obtained from a qualitative study that provide valuable insight into how to effectively design future collaborative tabletop file system interfaces.

2. Background and Related Work

Tabletop and pen-based interface research has explored interaction with small collections of information, such as digital photographs [3], or a desktop of files [1]. Research on importing files to tabletops has been limited. MultiSpace [8] enables explicit portal-based transfer of files between devices. The focus of this work is not on interacting directly with file systems from the tabletop, but interacting with specific files that have been explicitly and visibly transferred to the tabletop first. Other approaches [12, 16] require users to move their files from a laptop computer or personal storage device into a shared region of the tabletop display (and so, the user is privately copying files onto the tabletop). Furthermore, files are presented on the table in a flat unstructured collection. These systems do not support people interacting with broader collections of personal files at a tabletop.

Hierarchical tabletop interaction has been explored in the Personal Digital Historian project [18], as a way of increasing the scalability of a tabletop interface. However, hierarchical presentation tends to quickly fill the tabletop display, which makes navigation of multiple file system hierarchies problematic. Interaction with large hierarchies through passive and active associations has also been explored [17], although evaluation participants highlighted "clutter" and "over crowding" within groups of information as major problems in the interface.

The limits of hierarchies for managing files have been recognised as a problem in modern personal computing [6]. Various novel interfaces have been explored in an effort to improve the flexibility and efficiency of file systems. Haystack [13] allows users to create arbitrary relationships within their stored information to simplify retrieval, and the WinFS project is aiming to allow definition of relationships at the file system level [15]. Content-based desktop search (such as with Google Desktop [10]) allows interaction with a single information-space regardless of where information is stored. Associative file systems, such as the Semantic file system [9], also allow retrieval based on extracted file meta-data. However, these content-based access mechanisms provide targeted search interfaces that would be difficult to adapt to a collaborative tabletop, where there is a need to support concurrent interaction with multiple file collections.

3. Approach

Our approach is to develop and evaluate a new file access technique that simplifies retrieval of similar information from multiple file systems. While users have a strong existing mental model of hierarchical file system interaction, we need to explore new ways to interact with files that better leverage the possibilities afforded by collaborative interaction media. When designing for the tabletop, one needs to take account of the constraints described in the introduction: limited resolution; orientation independence; restricted modes of input; and problems with clutter. In addition, we have a key goal to support collaboration based on files from multiple file systems.

The design of our approach is to allow file access by similarity. Existing single-user file system interfaces provide hierarchical file browsers. We aim to support similarity-based file browsing to help people share relevant information when working together at a tabletop. The similarity-based browsing technique is based on the notion of finding 'similar' files (determined by file meta-data and text content) to a particular file of interest to the user(s). The system retrieves similar information to support the collaboration, allowing people to interact with useful information without needing to manually retrieve it first from different locations in multiple hierarchies. This allows associative file system interaction that is independent of the underlying storage structure, and the tools used to manage the information. However, we must provide flexible user control over the way that similarity is determined, as people organise and retrieve their files in different ways.

Another key element of our design is that the interface should preserve spatial layout when presenting files, and leverage the physicality of the tabletop. Our design avoids clutter in all aspects of the interface: in conjunction with the similarity-based browsing technique, files should be sized proportional to their similarity, so that most similar files are large and obvious to identify, while less similar files are small (though still identifiable) and take up less space on the tabletop. In addition, the interface must support concurrent multi-user interaction, both with the files on the tabletop and the navigation of the file systems.

This approach is unique in that the issue of enabling people to collaborate with their file systems at a tabletop has not been explored. Our design represents a new way to interact with files that is inherently collaborative — the design treats multiple people's file systems as a single information space, allowing users to interact with multiple file systems concurrently, without knowledge of where files are stored in the file hierarchies. This approach is significantly different from the single-user file system techniques used in current operating systems, but has the potential to address the constraints of multi-user tabletop interaction. We now present Focus, a tabletop interface that follows this design to support collaborative file system access.

4. Focus

Figure 1: People using Focus to share and discuss their personal file systems (samples shown left and right), with an architecture diagram showing their laptop computers connected to a central computer powering the tabletop display (bottom).

Figure 1: People using Focus to share and discuss their personal file systems (samples shown left and right), with an architecture diagram showing their laptop computers connected to a central computer powering the tabletop display (bottom).
Figure 2: The Focus interface showing files from multiple file systems, presented according to their relevance to the focus item (a text file about the Trojan War). The History Widget (top middle) shows thumbnails of past focus files to allow 'back' and 'forward' navigation. Dwelling on the circular icon returns to the start view of the file systems. The Black Hole (bottom right), allows users to hide unwanted files.

Figure 2: The Focus interface showing files from multiple file systems, presented according to their relevance to the focus item (a text file about the Trojan War). The History Widget (top middle) shows thumbnails of past focus files to allow 'back' and 'forward' navigation. Dwelling on the circular icon returns to the start view of the file systems. The Black Hole (bottom right), allows users to hide unwanted files.

Focus is a novel tabletop interface for collaboratively accessing multiple remote file systems. Focus (shown in Figure 1; see [5] for video) allows associative, similarity-based file access where a user focuses on one file of interest, with the system automatically retrieving all 'similar' files across multiple remote file systems, and showing only those deemed to be relevant. Focus has been developed as a plug-in to the Cruiser tabletop platform [2], a multi-user, gestural tabletop system that uses the Mimio Capture whiteboard pen system. Focus shows image representations of files that appear to be placed on top of the tabletop surface. Files can be moved by dragging, and they have realistic momentum, making it easy to flick files to areas of the tabletop that are out of physical reach. Files can be rotated and resized (in a combined rosize gesture) by selecting the file at one of its corners and dragging the corner. Files can also be flipped to see their filename written on the back.

When Focus is first launched, a broad start view of the file systems is presented in a radial layout on the tabletop. The start view shows the first file (alphabetically) in each exported directory of each remote file system. Navigation of file systems from the initial start view is based on the notion of a focus file. Once a user selects a focus file, all other similar files are displayed on the tabletop (regardless of where they are stored or which computer system and user they belong to). The file systems of multiple people are merged to appear as a single, combined information set on the tabletop, and a user may navigate the available information by re-selecting focus files. To select a focus file, a user dwells (depresses the pen for one second) on it. Matching results from each remote computer are returned and displayed immediately on the tabletop (see bottom of Figure 1).

Figure 2 shows Focus after two focus file selections from the start view. To address the limited display area of the tabletop, information is presented in a non-uniform manner, where the initial size of each file's image representation is determined by its relevance to the focus file. Thus, most relevant items appear large and prominent, while less relevant items appear small and unobtrusive. The resize gesture can be used to override the initial size based on similarity. If a file was previously displayed on the tabletop, it is presented in its last location if the user moved it to a specific position. This is to provide consistency between focus selections, and to allow users to create spatial groupings of their files.

A critical element of Focus is calculating file similarity. For the similarity-based browsing technique to operate effectively, it is essential that Focus will correctly retrieve the relevant files. Given the personal nature of file system organisation and content, we have created a mechanism for users to customise (with the Focus Exporter tool) how their files are determined relevant to a given focus. This is based on assigning weightings to 11 supported meta-data attributes (including the full text document content) in a simple user interface. Focus does not require users to manually attach meta-data to their documents, although keywords can be used to influence the relevance calculation. Since this paper concerns the interaction at the tabletop, we do not deal with the relevance calculation details, which are described elsewhere [4]. Focus Exporter uses the Apple Spotlight search framework, which scales well to large file systems. With the Focus Exporter running on an average desktop computer, results start to be displayed on the table within 2 seconds of a focus selection.

5. Evaluation

Figure 3: The Browser, a hierarchical tabletop file interface. The circled sub-folder (top left) has been dragged-off, resulting in a new folder object being created that shows the sub-folder's content (bottom left), which is a collection of image files. The circled file has been dragged to right to view it at original size.

Figure 3: The Browser, a hierarchical tabletop browsing interface. The circled sub-folder (top left) has been dragged-off, resulting in a new folder object being created that shows the sub-folder's content (bottom left), which is a collection of image files. The circled file has been dragged to right to view it at original size.

The goal of our evaluation was to gain insight into how tabletop file system interfaces are used for collaboration under realistic conditions, in order to inform the design of future tabletop file system interaction techniques. As this is the first work on tabletop file system access, there is a need for qualitative usage data. Thus, we conducted a qualitative study where we compared Focus to a hierarchical tabletop file interface called the Browser, an existing Cruiser plug-in typically used for browsing hierarchical photo collections. The Browser (shown in Figure 3) is a spatial file browser, where each folder in the file system is represented by a distinct object on the tabletop. This approach leverages the physicality of the tabletop, as each folder is represented as a realistic object on the table, and multiple folders can be open on the tabletop at once — this is critical for supporting concurrent access to multiple file systems. Folders and files can be opened by dragging the thumbnail off of the folder (a smooth tabletop gesture that is more resilient to accidental touches than 'clicking'), resulting in the file being loaded onto the tabletop, or the selected folder being opened. By comparing Focus with a hierarchical access interface implemented in Cruiser (instead of comparing with an existing single-user file system interface not designed for tabletop interaction, such as Windows Explorer), we ensure that the only variable between the conditions is the file system navigation technique, and not other aspects of the interfaces.

We recruited 10 participants (4 female) between the ages of 22 and 50, with varying backgrounds. Three participants had used a tabletop interface before (for less than 20 minutes each). All participants used a conventional hierarchical file system on a daily basis. The evaluation was conducted in pairs, and participants knew their partner beforehand. Participants completed five collaborative tasks that involved working together to create a set of information about ancient Greek history. One participant was allocated text documents related to the project, while their partner had photographs and e-mails. The tasks were designed to be authentic for collocated collaboration, representative of typical interactions with a file system, and to involve retrieving files from a diverse range of locations. A fictitious file system was used in the experiment to ensure comparability between trials.

To gain insight into the impact of file system size for each of the interfaces, two sets of test data were used: one having 59 files, the second having 114 files with more, and deeper, sub-directories. The test data represented a realistic subset of a file system. Keywords were assigned to images in order to enhance the relevance calculation performed by Focus, as the content-based search could not be fully utilised. A sheet with a personal file list, showing the hierarchical organisation scheme for the files they had been assigned, was given to each participant to refer to during the experiment.

The ordering of both the conditions used, and file system size varied between experiments. In the first step of the experiment participants were given a brief tutorial on the Cruiser interface. Participants were then given five minutes to practice manipulating sample images, followed by using each of the two file access conditions. After completing the tasks, participants were asked to complete a short questionnaire about computer use and feedback on each of the experimental conditions. The evaluation was captured on video, and the experimenter later performed a detailed analysis to identify interesting features of interaction and to code these for each condition.

6. Results

Figure 4: Questionnaire responses for ease of use, efficiency, and whether tasks were harder and slower when using the larger file system.

Figure 4: Questionnaire responses for ease of use, efficiency, and whether tasks were harder and slower when using the larger file system.

When using Focus, we observed participants employing a variety of techniques for navigating the file systems at the tabletop. Text documents (containing text about what participants were looking for) were used extensively as focus selections. The History Widget was also used extensively by all participants to go back to files that may be relevant to the task they were completing. We saw participants being social and co-operative when using this interface — eight participants suggested potential focus files in their own collection of files to help their partner find some in theirs. Participants exhibited a low sense of ownership of the file systems, as participants often interacted with and focused on their partner's files, with no perceived social boundaries inhibiting them from doing so.

When using the Browser, a significant issue was clutter, with participants adopting strategies to manage it (such as using designated corners of the tabletop for grouping files and folders, or placing them in the Black Hole). Four pairs informally adopted a clean-up protocol, where a participant would remove unneeded files and folders from the workspace before their partner began the next task. Notably, all participants were private with their file system interactions due to a higher sense of ownership when using the Browser. Participants would locate required files privately and then selectively share information with their partner.

Questionnaires indicate that participants found Focus consistently easier to use, and eight participants considered Focus to be more efficient, as shown in Figure 4. No participant found Focus harder or slower to use when the file system size was doubled, compared to seven who thought this was true of the Browser. All participants said the Browser interface had a familiar mental model, which initially made it easier for them to understand. Three participants expressed concerns about the similarity-based browsing technique failing to locate what they need. Two (3A, 4B) stated that they would like to see Focus include some features common to hierarchical file browsers, such as the ability to quickly navigate to specific files or directories, which is why they did not rate the efficiency of Focus highly (shown in Figure 4). For this reason, they commended the Browser interface because they can always access specific files.

All participants found Focus highly efficient at finding documents related to something that they were already working on, particularly when each file was in a completely different file hierarchy. This was highlighted by participants as an important feature for collaborating with team members over shared documents. One participant reported finding Focus very efficient to locate files because "it skipped through the hierarchy".

The Browser was received positively when compared with a conventional desktop computer file system browser due to its spatial nature. Three participants liked the fact that they could place as many folders as necessary on the table (and leave important ones open at all times), and it was easy to see the contents of a folder through the use of thumbnails rather than text.

7. Conclusions and Contributions

The most striking outcomes of the evaluation relate to the issues of collaboration and management of clutter in tabletop interfaces. Beginning with collaboration (a key goal for table interaction), we saw that Focus provided valuable support for tabletop collaboration when working with personal file systems. People were able to collaborate more on the file access process, rather than only selectively sharing information after it had been found with the Browser. However, the Browser provides hierarchical access when needed, which may be required in cases where the similarity-based browsing technique fails, or when users would prefer to explicitly access certain files based on their storage location. The hierarchical paradigm of file system interaction encouraged people to work privately when locating relevant information, and resulted in people creating clutter that impacted on other users of the tabletop. In some cases, participants did not remove unneeded objects from the tabletop to clear space for their partner, causing difficulties using the interface due to clutter. In other cases, participants adopted strategies to manage clutter, requiring more effort when locating files. Research of multi-user clutter management techniques is a clear area of future tabletop research.

The browsing technique used in Focus was received positively, despite users having a strong existing mental model of hierarchical file system interaction. With Focus presenting an amalgam of the relevant files from the users' file spaces, we observed participants collaborating more and helping each other to find the required information. The automatic clutter management in Focus (by only showing files similar to the focus file) was well received, and prevented the interface from becoming noticeably cluttered (even when the file system size was doubled in the evaluation). Our evaluation highlights the need to continue exploring non-hierarchical file system interfaces that enhance multi-user collaboration at a tabletop. While we have applied the similarity-based browsing technique to a tabletop interface, this technique could be applied to a variety of media, including conventional desktop file system interfaces to allow efficient retrieval of similar files (possibly stored on multiple machines belonging to the same person).

The key contributions of this work are:

Acknowledgements

Thank you to my advisor, Associate Professor Judy Kay, for her guidance and feedback. The software prototype created in this project was made possible by the plug-in architecture of Cruiser developed by Trent Apted. Thank you also to Alan Fekete and Kalina Yacef for their feedback on this paper. This work is partly funded by the Smart Internet Technology CRC.

References

  1. A. Agarawala and R. Balakrishnan. Keepin' it real: Pushing the desktop metaphor with physics, piles and the pen. In Proc. CHI 2006, pages 1283–1292, New York, NY, USA, 2006. ACM Press.
  2. T. Apted, J. Kay, and M. Assad. Sharing digital media on collaborative tables and displays. In Proc. The Spaces In-between (UbiComp 2005 Workshop), 2005.
  3. T. Apted, J. Kay, and A. Quigley. Tabletop sharing of digital photographs for the elderly. In Proc. CHI 2006, pages 781–790, New York, NY, USA, 2006. ACM Press.
  4. A. Collins. Exploring tabletop file system interaction. Undergraduate Thesis, School of IT, University of Sydney, http://www.it.usyd.edu.au/~anthony/thesis.pdf, November 2006.
  5. A. Collins. OnTop: A collaborative tabletop file system interface, (video of the Focus interface), http://youtube.com/watch?v=2KKABHIpKzM, 2007. Retrieved 29 April 2008.
  6. P. Dourish, W. K. Edwards, A. LaMarca, J. Lamping, K. Petersen, M. Salisbury, D. B. Terry, and J. Thornton. Extending document management systems with user-specific active properties. ACM Trans. Inf. Syst., 18(2):140–170, 2000.
  7. P. Dourish, W. K. Edwards, A. LaMarca, and M. Salisbury. Presto: an experimental architecture for fluid interactive document spaces. ACM Trans. Comput.-Hum. Interact., 6(2):133–161, 1999.
  8. K. Everitt, C. Shen, K. Ryall, and C. Forlines. MultiSpace: Enabling electronic document micro-mobility in table-centric, multi-device environments. In Proc. TABLETOP 2006, pages 27–34, Washington, DC, USA, 2006. IEEE Computer Society.
  9. D. K. Gifford, P. Jouvelot, M. A. Sheldon, and J. W. O'Toole Jr. Semantic file systems. In Proc. ACM SOSP 1991, pages 16–25, New York, NY, USA, 1991. ACM Press.
  10. Google desktop, http://desktop.google.com, 2008. Retrieved 29 April 2008.
  11. Google search basics: Similar pages, http://www.google.com/support/bin/answer.py?answer=49225, 2008. Retrieved 25 April 2008.
  12. B. Hartmann, M. R. Morris, and A. Cassanego. Reducing clutter on tabletop groupware systems with tangible drawers. In Adj. Proc. UbiComp 2006, 2006.
  13. D. R. Karger, K. Bakshi, D. Huynh, D. Quan, and V. Sinha. Haystack: A customizable general-purpose information management tool for end users of semistructured data. In Proc. CIDR 2005, pages 13–27, 2005.
  14. D. M. Ritchie and K. Thompson. The UNIX time-sharing system. Commun. ACM, 17(7):365–375, 1974.
  15. T. Rizzo. WinFS 101: Introducting the new windows file system, http://msdn2.microsoft.com/en-us/library/aa480687.aspx, 2004. Retrieved 29 April 2008.
  16. C. Shen, K. Everitt, and K. Ryall. UbiTable: Impromptu face-to-face collaboration on horizontal interactive surfaces. In Proc. UbiComp 2003, volume 2864 of LNCS, pages 281–288, Berlin/Heidelberg, 2003. Springer.
  17. C. Shen, N. B. Lesh, F. Vernier, C. Forlines, and J. Frost. Sharing and building digital group histories. In Proc. CSCW '02, pages 324–333, New York, NY, USA, 2002. ACM Press.
  18. F. Vernier, N. Lesh, and C. Shen. Visualization techniques for circular tabletop interfaces. Technical Report TR2002-001, Mitsubishi Electric Research Laboratories, March 2002.