WWW3D is a novel web browser that provides a 3D display which integrates the display of web pages and structural information about the portion of the web that the user has explored. This is achieved by using a novel interpretation of HTML which produces 3D representations of web pages instead of the more normal 2D layout provided by normal web browsers. In addition to this WWW3D supports multiple simultaneous users who may be browsing quite different areas of the web. In this paper we describe the ideas behind WWW3D, the prototype implementation and discuss how this work can be extended.
Most current web browsers display only the representation of the web document(s) that the user is currently examining with no indication of the structure of the document other than by highlighting hypertext links. Those browsers that do indicate the structure of web documents, such as the Harmony browser for Hyper-G [Hyper-G], use a different window from that of the document to present this information to the user. The documents that the user has examined are likely to be displayed in yet another window or as a menu.
In this paper we will describe WWW3D a novel browser, implemented in DIVE [Hagsand'96, Carlsson'93], that provides a single 3D display which integrates the display of the web documents themselves, the structure of the part of the web that the user has browsed and history information showing the links the user has followed in the recent past. In addition to this WWW3D supports multiple concurrent users who are visible to each other and who may either be browsing the same or different sets of web documents.
Sections two and three of this paper will describe the techniques used to represent web documents and the means by which the structure of the web is displayed to the user. Section four will describe some techniques that we plan to apply in order to produce a more functional version of WWW3D. Finally, we shall end with a discussion of the usefulness of WWW3D and of possible improvements that could be made to it.
Many HTML tags provide semantic information about the marked-up text - i.e. what the text is rather than how it should be displayed on the page. 2D web browsers will typically use this information to determine what style (i.e. font, font style, font size, etc.) to use when drawing a particular piece of text.
WWW3D uses the information contained in HTML tags to produce a representation of the document in 3D space. A web document is represented as a sphere which is labelled with the document's title. The contents of the document is placed around the inside surface of the sphere. Displaying large amounts of text in a satisfactory way is difficult in current VR systems so textual information is currently represented by icons that can be unfolded to reveal the entire text. The first few words of the piece of text are displayed under the icon to give some indication of the contents. Images are displayed by texture mapping them onto polygons on the inside surface of the sphere. Finally, links to other documents are represented as icons labelled with their destination.
To reduce the visual complexity of the virtual environment WWW3D makes extensive use of Level of Detail (LOD) operations. When viewed from outside a document is represented as an opaque sphere and the actual document contents is not displayed. When a user enter's a document to view it the sphere is drawn in wireframe so that the rest of the virtual environment is still visible.
Figure1. A WWW3D representation of a web document seen from the inside
.
Figure 1 shows the contents of a web document as displayed by WWW3D. Figure 2 shows the display generated when a user reads some of the text comprising a web document
When the user selects a link icon, WWW3D creates a new sphere representing the target document and places it near the document from which the user selected the link. In order to indicate the structure of the portion of the Web that the user has explored WWW3D draws arrows between the spheres representing linked documents. If the documents are resident on the same Web server then the arrow is drawn in blue, otherwise it is drawn in green thereby helping to provide additional information on the structure of the documents that the user has explored. In addition to this, the brightness of the arrow is dependent on the time since the user last followed that link thereby providing the user with a visual representation of their browsing history. If WWW3D fails to fetch a document then a small red arrow is attached to the source document to represent the "broken" link.
WWW3D is implemented using multiple lightweight threads so the user does not have to wait for a document to be retrieved before selecting another link. This behaviour is essential if multiple users are to be able to browse independently. Users are also free to navigate through the space and browse other documents while waiting for a document to be retrieved.
As WWW3D parses a newly retrieved document, it checks for links to documents that the user has already explored and draws arrows to represent them. This means that at any given moment the complete set of links between documents is displayed without the user having to follow every link. This is intended to aid the user by indicating links between documents that the user might have been unaware of. This also has the result that several users can be browsing different parts of the web and yet any links between the sets of documents they are exploring will be displayed. This might be useful since users will then have a visual representation of possible common interests.
To produce an acceptable layout of the set of linked documents an incremental version of the Force Directed Placement (FDP) [Fruchterman'91] algorithm is used. Links between documents act like spring forces which result in linked documents being moved closer together. Documents exert repulsive forces on one another which prevents documents being placed closer together than a user-specified minimum separation.
At regular (user specified) intervals WWW3D applies the FDP algorithm to refine the inter-document layout. The more links a document has, the greater "inertia" it is considered to have when the FDP algorithm is applied. This has the result that heavily referenced documents are less likely to move and provides some stability to the visualization. In addition damping is applied to try and prevent large changes to the visualization for a given iteration of the FDP algorithm - this helps to prevent the user from becoming disorientated. The result of this is that the inter-document layout gradually evolves over time to produce clusters of inter-linked documents.
This incremental use of the FDP algorithm is similar to the Narcissis system [Hendley'95]. However, Narcissis uses a separate web browser to display the contents of web documents and does not support multiple users.
Figure 3 shows the display produced by WWW3D after the user has browsed a number of documents. The FDP algorithm has resulted in the formation of clusters of closely linked documents. The colours of the documents provide some indication of how long ago the user last visited them.
Figure 3. A collection of web documents, showing the links drawn between them.
Between invocations WWW3D stores information on the current set of documents, the links between them and the current 3D layout to a file. When a new instance of WWW3D is started, it reads this history file and displays the structure as it appeared in the last session. The contents of the documents found from the history file are not retrieved until the user enters a particular document. Since the document representations are opaque this process is invisible to the user except for an occasional delay in seeing a document's contents.
Thus far we have described our initial prototype which we built using the SICS DIVE virtual environment. Based on our experiences with the current prototype we have identified several areas where improvements could be made; these include improving the layout of HTML in 3D, increasing the scalability of the system so it can handle more complex visualizations, providing better support for searching through the visualization and representing other web users. This section will briefly describe these planned improvements and describe the key features of MASSIVE-2 the virtual environment that we are using for WWW3D version 2.
The Communications Research Group at the University of Nottingham have recently implemented a new collaborative virtual environment called MASSIVE 2 [Benford'97]. MASSIVE 2 provides some novel functionality including:
Using MASSIVE 2 to implement the next version of WWW3D will allow us to re-implement WWW3D in an object-oriented manner and exploit MASSIVE-2's ability to handle hierarchic multicast groups to improve scalability.
A current problem with WWW3D is that although it makes extensive use of Level-of-Detail operations for individual nodes, it will still eventually reach a point where the world becomes too complex to be rendered on even top of the range hardware. This could be avoided by clustering the visualization and representing clusters of nodes as single artifacts that are only expanded into individual artifacts when the user becomes sufficiently aware of them. If clusters can be composed hierarchically, then the application will be far more scalable since arbitrarily complex clusters of nodes can be represented as single artifacts providing they are below a specified awareness threshold.
MASSIVE 2 allows a virtual world to be decomposed into regions each of which may have its own associated IP-multicast group. These regions can be constructed so that only members of a region are sent state information about other members of the region. Therefore by representing clusters of nodes as MASSIVE 2 regions we would not only reduce the number of artifacts that must be rendered, but also reduce the amount of information that must be distributed to members of the world, thereby reducing the required network bandwidth and increasing the overall scalability.
In order to do this we need a mechanism for grouping nodes into clusters. LEADS [Ingram'95] is a system that attempts to make visualizations of abstract information spaces more legible by grouping relating artefacts into clusters and by adding other features to aid user navigation such as landmarks. HyperVisVR [Brown'96] uses a hierarchic clustering method. Both systems have already been applied to visualizations produced by the FDP algorithm (as used in WWW3D) and so are applicable to clustering WWW3D visualizations.
Currently, WWW3D only fetches new pages when a user selects a hypertext link and so its behaviour is equivalent to a typical web browser. However, there are other possibilities that could be implemented to provide additional features, for example:
Another useful extension would be the provision of support for searching the web. WWW3D currently explicitly shows structural information about web documents, but semantic information about the contents of web pages is only displayed when a user visits a particular web page. It is therefore hard to find information about a particular subject without having to manually browse the web pages. One possibility for improving this would be to add more semantically based visualization features that support searching for information. An example of this style of visualization is VR-VIBE [Benford'95] which allows users to browse bibliographic information by selecting an arbitrary number of keywords which are used to control the layout of document icons in 3D space.
Currently the only web users visible to users of WWW3D are those using WWW3D itself. Representing other users is important since it allows users to collaborate and provides for the possibility of chance meetings. However, the vast majority of web users use traditional browsers and are not currently represented in the WWW3D space. It might be useful to be able to represent all web users in the WWW3D space for several reasons, including:
There are several possible techniques for tracking web users and finding which pages they are browsing, Brown [Brown'96] discusses several of these including server extensions and the use of Java applets to send information about the user to a logging service.
Obviously, if we attempt to display representations of many thousands of individual web users we run the risk of overloading the system. However, research into the use of aggregate views of large collections of users may be applicable here [Benford'97]. These techniques are based on the idea of identifying groups of users and using a single artefact to represent the group instead of displaying each user individually. Members of a particular group may (optionally) see individual users rather than the aggregate view.
WWW3D is a only at a prototype stage of implementation and still requires several refinements. In particular the layout of document contents still requires work in order to produce an easily readable display for documents that contain large amounts of text. In addition general formatting of text could be improved.
Since WWW3D was originally designed for use immersively or in a CAVE [Cruz'92] type display all information is displayed in a single 3D window. For desktop use it would be possible to use separate 2D windows to display text if the user preferred.
User navigation could also be better supported in WWW3D. Currently, a user must manually navigate between document representations. Additional navigation support is planned which would allow users to select the arrows representing inter-document links causing them to be smoothly moved towards the target document.
Finally, although the authors have found WWW3D quite easy to use and quickly became accustomed to the way is which web documents are represented more extensive user trials are required to discover whether users will accept such a radical departure from the normal methods of displaying the contents of web documents.
In this paper we have described WWW3D, a novel Web browser that integrates information about the contents of web documents, the structure of the portion of the Web explored by the user and history information. WWW3D is implemented using the SICS DIVE VR system and can support multiple concurrent users. Users may communicate with each other using standard facilities provided by DIVE. We also described our plans for a new version of WWW3D to be implemented in MASSIVE 2.This new versions will use clusters and regions to improve scalability, have added support for searching and will be capable of displaying representation of the activity of the users of traditional web browsers.
[Benford'95] Benford, S., Snowdon, D., Greenhalgh, C., Inrgam, R., Knox, I., and Brown, C. (1995), "VR-VIBE: A Virtual Environment for Co-operative Information Retrieval", Computer Graphics Forum 14(3) (proceedings of Eurographics'95), 30th August - 1st September, NCC Blackwell, pp 349-360.
[Benford'97] Benford, S., Greenhalgh, C., Lloyd, D. (1997): "Crowded Collaborative Virtual Environments", to be presented at ACM CHI'97. (in press).
[Brown'96] Brown, C., and Benford, S. (1996), "Tracking WWW Users: Experience From The Design of HyperVisVR" Proc. WebNet'96, San Francisco, October 15-19, 1996, Association of the Advancement of Computing in Education (AACE).
[Carlsson'93] Christer Carlsson and Olof Hagsand, "DIVE - a Platform for Multi-user Virtual Environments", Computers and Graphics, 17(6), 1993, pgs 663-669.
[Cruz'92] Carolina Cruz-Neira, Daniel J. Sandin, Thomas A. DeFanti, Robert V. Kenyon and John C. Hart, "The Cave - Audio Visual Experience Automatic Virtual Environment", Commnuications of the ACM, 35(6), June 1992, pgs 65-72.
[Fruchterman'91] T.M.J. Fruchterman and E.M. Reingold, "Graph Drawing by Force-directed Placement", Software Practice and Experience, 21(11), Nov. 1991, pgs 1129-1164.
[Hagsand'96] Olof Hagsand, "Interactive Multiuser VEs in the DIVE System", IEEE Multimedia, 3(1), 1996, pgs 30-39.
[Hendley'95] R.J. Hendley, N.S. Drew, A.M. Wood and R. Beale, "Narcissus: Visualising Information", Information Visualization'95, IEEE Computer Society Press October 1995, Atlanta, Georgia, USA, pgs 90-96.
[Ingram'95] Ingram, R., and Benford, S., (1995), "Legibility Enhancement for Information Visualisation", in Proceedings of Visualization'95, Atlanta, Georgia, November 1995
[Hyper-G] http://www.tu-graz.ac.at/about