Proceedings of the 2nd Workshop on Adaptive Systems and User Modeling on the WWW
The 8th World-Wide Web Conference, Toronto, CA, May 11, 1998

Capturing Interaction Histories on the Web

Robert Farrell
IBM T J Watson Research Center
Applied Learning Sciences Group
Yorktown Heights, NY 10598
phone: (914)945-3398 / fax: (914)945-4395
Abstract: The World-Wide Web offers the unprecedented ability to reach large numbers of users with a single application. While some users may wish to customize their applications, in many situations it will be desirable for applications to adapt themselves to users. An important step toward adaptive systems is providing a way of capturing the history of interactions between  users and the applications they use. The current work looks at ways of capturing and interpreting both user inputs and application responses. These interaction histories can be shared between clients, clients and servers, or between servers on the World-Web Web.

1. Introduction

The World-Wide Web (WWW) offers the unprecedented ability to reach large numbers of users with a single application working over a wide area. For such applications to be truly useful to such a wide audience, they must be customized. However, the large number and changing distribution of users on the Web makes it virtually impossible to customize applications manually for each user. While applications may be customized by users themselves, there are many situations where it would be desirable for applications to adapt themselves to the users, instead of the other way around [1].

In educational settings, students struggle with slow and complex Web interfaces designed for adults. Adaptive interfaces could make access to the wealth of information on the Web easier for school children without requiring them to go through a complex customization before starting to see benefits. As electronic commerce becomes more ubiquitous on the Web, new users are spending much of their time trying to find the product they need. Adaptive interfaces could make the path to products easier for the inexperienced while not requiring casual users to spend time customizing an interface they may use only once.

2. Interaction Histories

Web-based applications need to take into account not only stated interests and preferences, but actual history of use. Toward this end, we need standards for encoding usage history so that data can be safely and easily shared between clients and other clients, clients and servers, and servers and other servers on the WWW. Better ways to process user history across Web sites and systems could have wide application. Web sites could redesign themselves to suit their changing user population, help desks could better understand what happened before a service or support request [Farrell et al, 1997], and training systems could provide more user-centered task guidance [Farrell & Lefkowitz, 1998].

At T J Watson Research Center, we are working on methods of capturing the interaction between users and applications by extending the operating system (OS) desktop. Our approach is to interject a layer of processing between applications and the user that captures both user inputs and system responses. Our system filters, abstracts, and correlates OS-level events, drawing  inferences linking observed event patterns with postulated user goals. We believe that these methods will apply to the Web.

To capture interaction histories on the Web, we must be able to capture not only hyperlink selections and HTML form use, but also plug-ins, applets,
and client-side scripts created with Javascript, Java, and other languages. We propose that, upon launching a program or component of any type, the browser or other Web client request that the operating system perform interaction monitoring. Results from monitoring are reported back to the Web client in an operating-system independent way. Interaction histories stored by the Web client would initially be private, much like Web access histories. However, we would provide a mechanism for a server to request the access history of the current. This mechanism would allow applications to retrieve and process the history of interactions with a given user, an important first step in adapting to that users' needs.

3. Research Questions

While proposals exist for sharing interest profiles and demographic data across Web sites [Netscape, 1997][Microsoft, 1997], we are concerned here with capturing and sharing a potentially much richer source of data. However, many questions need to be answered if we are to take the approach advocated here. How can interaction data be kept private [W3C-P3P, 1998] and secure? How can interaction data by represented so that it can be transferred between applications? What kind of performance problems may arise if this kind of data is captured and transmitted across wide area networks?


[Brusilovsky et al, 1997] Brusilovsky, P., A. Kobsa and J. Vassileva, eds. (1997): Adaptive Hypertext and Hypermedia Systems. Dordrecht, Netherlands: Kluwer Academic Publishers.

[Farrell et al, 1997] Farrell, R., Kukich, K., and Jones, Frank T. Publishing Troubleshooting Expertise on the World-Wide Web: A Case Study. Proceedings of the WebNet Society Conference, 1997.

[Farrell & Lefkowitz, 1998] Farrell, R. and Lefkowitz, L. "Supporting Development of On-line Task Guidance for Software System Users." In Facilitating the Development and Use of Interactive Learning Environments, C. P. Bloom & R.B. Loftin, (Eds.), 1998.

[Netscape, 1997] Netscape Communications Corporation. "Open Profiling Specification (OPS)", Web Consortium Submission #6 (at ssion/1997/6/Overview.html.) Jun 2, 1997.

[Microsoft, 1997] Microsoft Corporation. "Privacy and Profiling on the Web", Web Consortium Submission #7 (at ssion/1997/7/Overview.html.) Jun 2, 1997.

[W3C-P3P, 1998] World-Wide Web Consortium (W3C) "Platform for Privacy Preferences (P3P1.0) Specification". Working Draft #WD-P3P-19981109 (at Nov, 9, 1998.