Proceedings of the 2nd Workshop on Adaptive Hypertext and Hypermedia,
HYPERTEXT'98, Pittsburgh, USA, June 20-24, 1998

Computed Web Links: The COOL Link Model

Michael Miller and L. Jay Wantz
Intelligent Automation, Inc.
Rockville, MD 20850 USA
{mm@i-a-i.com, jwantz@i-a-i.com}
phone: (301)590-3155, fax: (301)590-9414

Abstract: As the Web evolves it is likely to become more user-centered, where navigation and search are more personalized and individualized. Signs of this are already surfacing in various guises including filters and other censorship tools, the use of cookies as user profile repositories, and metadata-aided search. Whether the user-centered approach will become the predominant model is open to debate, but it does appear that it will continue to make inroads. With minimal adaptations the technology available today can better manage and deliver the vast amount of information on the Web by meeting the user's needs, goals, and preferences. This paper discuss one such adaptation: COOL links. These are multi-ended, computed hyperlinks, which foster a new model of Web navigation, applicable to a variety of disparate domains, including education and marketing.
Keywords: multi-ended hyperlink, computed hyperlink, user-centered navigation, individualization, user profiles

1. Introduction

Today's Web navigation features user manually selecting their own path through single-ended "hot" links. This user-directed approach, though successful and useful for many purposes, has a number of drawbacks, which, as more and more information becomes available on the Web, seem to loom larger. In particular, navigation is often punctuated by frequent and lengthy searches and dead-end paths through irrelevant, undesired, unsolicited or inappropriate information. The result is that the user is often unaware of relevant aspects of a collection of resources until he or she has navigated through much unwanted material. In addition, it is not unusual that clicking on a link leads to surprises, irrelevant paths, or a feeling of being lost in "cyberspace."

Traditional navigation ignores an individual's specific preferences, needs, or other abilities (e.g., reading level). This can be particularly problematic when the user is under time, or learning and comprehension constraints, as in educational situations. An alternative is automated guidance toward material that is of interest, contextually relevant, and appropriate for the user. In an educational setting it would be useful for a learner to be guided along a path of Web resources which meet his or her educational needs. In the commercial arena it would be useful to direct the Web user to those products and services in which he or she might be especially interested.

Notable examples of such non-traditional approaches are: Walden's Paths [FU97], WebWatcher [JO95] and Letizia [LI95]. These systems have shown some promise, especially in locating pages similar to those already found to be of interest to the user, but stop short of locating Web resources for specific users with specific educational needs and goals. With minimal adaptations and the technology available today, the vast amount of information on the Web can be even better managed and delivered to the meet the user's needs, goals, preferences and attributes. COOL links [WA97] are one such adaptation.

2. Not all Link Types are Created Equal

Most familiar to the Web community is the single-ended, uni-directional "hot" link, denoted by the familiar HTML href tag in an anchor block. Each link of this type is intended to refer to the address of exactly one resource, which itself is denoted statically in the HTML. Thus, one of the prime attributes and attractions of these links is that they are unambiguous. Another is that the HTML author knows exactly where a viewer of the page will "go" if a link is clicked.

Other link types, such as multi-ended links, do exist and have been implemented in a variety of other hypermedia environments (not the Web). At first it seems that a multi-ended link--one which "refers" to multiple resources--applied to the Web would be ambiguous. To which of the multiple resources is the user taken when the link is clicked?

COOL links are an example of multi-ended Web links that are not ambiguous, because a COOL Link is also a form of a computed link. This is accomplished by letting these links "refer" to a collection of resources only until click-time. The resources associated with a COOL link are unordered. There is no special significance given to those that appear closer to the start of the link description than those that appear nearer the end. Instead, when a COOL link is clicked the metadata of each component resource is evaluated and compared against a separate set of input parameters, for instance the user's profile, and the "best-fit" resource is returned.

Presumably, each distinct resource in a COOL link collection provides different benefits to different users (though each resource may contain information about the same subject or topic). The burden of choosing a link component from the collection is placed upon the browsing tool (or a plug-in) at runtime.

In summary, the four-part COOL link machinery contains: (i) The link itself (its implementation), (ii) metadata descriptions of the linked resources, (iii) external input features such as user profiles, and (iv) link computation algorithm(s).

2.1 Link Implementation

Our current implementation of COOL Links uses a Javascript function named CoolLink to create the computed nature of the link. The CoolLink function takes a variable number of arguments, where each argument represents the metadata of one of the destination resources in the multi-ended link. The metadata contains a pointer to the Web resource it describes, thus separating resource description and annotation from content authoring. In HTML, a COOL Link might look like the following:

<a href="default.html" OnMouseOver="this.href=
                       CoolLink('res1.rdf', 'res2.rdf', 'res3.rdf', 'res4.rdf')"></a>

A single resource is selected from the collection by an evaluation scheme, based upon the user's profile. In this implementation, a user must first log onto the system before the COOL Links will function properly. The user's login name is stored locally in a cookie, but the profile information is stored in a server-side database for security reasons. If no user has logged onto the system, or if a link cannot be computed, the browser destination is set to the default location specified in the href attribute.

Our implementation of COOL Links will be further enhanced in the future. With the acceptance of the XML specification, the W3C will be supporting the notion of multi-ended links on the Web. Therefore, part of the COOL Link machinery will be supported inherently by the markup language. Secondly, we will allow for different types of computational algorithms to be associated with a given COOL Link. A COOL Link computation doesn't necessarily have to involve a comparison with the user's profile, it might be based upon other external input. Finally, in cases where the user's profile is required, we will be caching the user profile locally, and eliminating additional server hits.

2.2 Resource Metadata

Today's Web community (including the W3C, the major browser makers, and the creators of the search engines) has embraced the collection and use of metadata to characterize Web resources. Several notable metadata efforts have made an impact, directly or otherwise, on the Web. Among those efforts, and of interest to our current work, are the Dublin Core [WE97], the Gateway to Educational Materials (GEM) [SU98], the Warwick Framework [DE96], the Learning Objects Metadata Group (LOMG), and the Instructional Management System (IMS).

In our past implementations of COOL Links, we used the Courseware Description Language (CDL) [HA96tr, WA98] metadata dictionary to describe Web resources. The CDL dictionary was first developed in late 1995, independent of the metadata work in digital cataloging (Dublin Core) and online clearinghouses (GEM). Yet, as one might suspect, there are many similarities (e.g., most elements of the Dublin Core map directly to elements in CDL). The CDL is an extensive collection of more than 50 attributes for describing educational resources that are particularly appropriate for K-12 education applications. Descriptions may characterize a resource's applicability to specific grade levels, its support of various pedagogical models, its time requirements, its dependency on various teacher roles, and its relationship to other resources in a set, to name a few. One of the primary goals in the development of the CDL was to address the individual needs of students, including the special needs of students with disabilities. Through the use of specialized tools, such as COOL Links, we now can construct educational hypertext on the Web that contains dynamic paths. As an individual student clicks a link, his user profile is compared to the CDL descriptions of the associated resources, changing the path to meet his needs.

In our current implementations we are adopting the W3C's Resource Description Framework (RDF) syntax for representing metadata on the Web. We have embraced the Dublin Core and GEM metadata dictionaries as core components, while adding extensions from our own CDL dictionary, to create resource descriptions that follow the concepts of the Warwick Framework. The following is an example of how an eighth grade resource in biology might be described using RDF syntax and the three metadata dictionaries in combination:

<?xml:namespace name="http://purl.org/metadata/dublin_core#" as="DC"?> 
<?xml:namespace name="http://www.geminfo.org/#" as="GEM"?>
<?xml:namespace name="http://www.i-a-i.com/services/cdl/#" as="CDL"?>
<?xml:namespace name="http://www.w3.org/TR/WD-rdf-syntax#" as="RDF"?>
<RDF:RDF>
   <RDF:Description RDF:HREF="http://www.i-a-i.com/LabExperiment.html"> 
      <DC:Creator>Olaf Kolzig</DC:Creator>
      <DC:Format>text/html</DC:Format> 
      <DC:Date>1998-21-05</DC:Date> 
      <DC:Title>A closer look at single-cell micro-organisms</DC:Title> 
      <GEM:Grade>8</GEM:Grade>
      <GEM:Pedagogy>Self-directed groups</GEM:Pedagogy>
      <GEM:Duration>Two one-hour sessions</GEM:Duration>
      <CDL:UserRole>Discovery</CDL:UserRole> 
      <CDL:UserMode>Multi-sensory</CDL:UserMode> 
      <CDL:Equipment>Microscope, slides, eye dropper</CDL:Equipment> 
   </RDF:Description>
</RDF:RDF>

2.3 User Profiles

A key concept here, and one that separates the current work from other Web-based metadata applications, is our employment of user profiling together with metadata to help locate Web resources. While traditional navigation ignores the specific traits of an individual, the user-centered approach relies on them. The characteristics of the user are compared with Web resources during search and navigation, helping to prune the information space and direct the user toward the most relevant resources.

Most of our current COOL link implementations are in the context of delivering educational material via the Web, and thus the class of profile traits is circumscribed by pedagogy. Clearly, however, this will need to be expanded for future applications. We do not see this as a problem, however. Extensive profiles, containing both domain dependent and independent traits may be maintained, though not all traits need factor in the comparisons done for a particular COOL link. (Exactly which traits are to be used are determined by context, and go into parameterizing a Selection Algorithm--see below.)

To date, we have used only manual profiling techniques, though we see this as insufficient. One problem with manual profiling is that the user is required to maintain the profile, making modifications as his or her characteristics change. For instance, as a student's reading abilities improve, the profile must be updated to reflect that progression. Otherwise, as a profile becomes out of date the user will be directed toward resources that are no longer relevant. Not only does this become obtrusive, but also there are traits and scales for which the user is likely unaware of precisely where he or she fits. Reading level, which may be based on a Flesch-Kincaid and/or Fog indices of readability, is an example.

In contrast, a (semi-)automated approach requires less direct interaction from the user in order to develop and maintain a comprehensive profile. There have been localized successes in the area of machine learning, where one or more preferences or characteristics can be inferred, but not a comprehensive user profile. The Web-based systems that have recently been built (including [JO95] and Letizia [LI95]) tend to focus on identifying those resources which will be of interest to a user, though not on identifying those which will meet the user's educational goals, given their capabilities and knowledge. We continue to investigate this fertile area of research.

2.4 Selection Algorithms

Resolving a COOL link, i.e., selecting a link from the collection of links contained in a COOL link, is the task of a selection algorithm contained in the browsing tool or a plug-in. This selection algorithm compares a resource's metadata characteristics with aspects of the user's profile, or other external features, such as time of day, domain name of the user's machine, etc., and outputs the resource found to be "most appropriate."

To facilitate this, metadata descriptions of each component of a COOL link must be fetched. In general, fetching is intended to be carried out at runtime, i.e., at the time that a COOL link is clicked, however, some optimization can be achieved (when a page contains a relatively small number COOL links) by doing all of the metadata fetches necessary for a page as a background process begun at page load time. Then when a COOL link is clicked, no fetch time is required.

Various criteria for choosing a component resource can be implemented by link selection algorithms. We have developed a generic link selection mechanism and incorporated it into our Web implementations (see, [WA97]), but other specialized algorithms, or parameterization of the generic version will be useful in future applications and contexts. We intend this selection component to help deal with some of the the domain dependence vs. domain independence issues mentioned above. Namely, the context of use of a page containing COOL links can help determine the parameters of a generic algorithm, or the specialized algorithm, which should be used for resolving the page's component COOL links. Thus, pages used in an educational setting could invoke a different selection algorithm than those used in a more commercial context. Note, however, that a single user profile could serve both, with profile traits being weighted in accordance with the domain.

3. Applications

COOL links are applicable in a variety of domains and in a variety of different ways, including:

Education: Students tend to progress at their own pace, learn differently, and have a diverse set of skills, even though they are following the same lesson. Ideally, an instructor takes a student's characteristics into account, and teaches accordingly. COOL links can serve as a tool to facilitate lesson individualization in an electronic learning environment. As students navigate Web-based lesson material templates, they can be guided on a path that is most appropriate to their learning goals and capabilities. A link selection mechanism determines a path dynamically as a student browses through courseware containing COOL links. An instructor or lesson developer can create a single lesson plan template which contains different material for use by students with differing needs. Attributes like the pedagogical "appropriateness," associated grade level, and readability index of a resource in relation to a student's profile may weigh heavily on the choices made by the selection mechanism.

User-centric Navigation: A great deal of research has been carried out on Web trails, paths and guided tours[TR88]. Trails tend to be contextually relevant and interesting to the user, but are not necessarily shaped by characteristics of the individual user himself. Taking these characteristics into account leads to user-centric navigation (UCN)--the automated guidance of an individual through the Web.[WA97] UCN is a paradigm shift from the more passive modes of hypertext navigation as it facilitates the development of dynamic trails based upon context, user interaction and user needs.

Advertising, Marketing and Sales: The color, size, cost, or style of a particular consumer item in relation to a consumer's preferences may affect his or her desire to purchase an item. Additionally, other attributes of the consumer like background, gender, age, career and income will affect the likelihood of a purchase. With advertising and sales a mainstay of the Web, sellers are looking for ways to direct those most interested in their products to their pages. COOL links, together with user profile data, are one way to accomplish some amount of successful marketing, and in the process eliminate large numbers of unwanted hits.

4. Conclusion

As the Web continues to grow, more and more information will become available. Experience has shown that the more information available, the harder it will be to find what we want unless some measures are taken to better tailor navigation and searching technologies. There will continue to be times that traditional user-directed Web navigation will be entertaining and informative, but there will be other times when it will be frustrating and wasteful. Multi-ended, computed COOL links offer a new way to view the Web, with minimal impact on the existing Web architecture. With the current emphasis placed on metadata and user profiling, and the mounting frustrations stemming from wasted time spent searching for resources, we hope that COOL links will continue to make inroads into the Web, especially as new applications highlight their benefit and exemplify their utility.

Acknowledgments

This research and development was supported (in part) by DARPA contract #N66001-95-C-8626, and US Department of Defense Education Activity (DoDEA) contract #N66001-97-M-C528.

References

[DE96] Dempsey, L., Weibel, S. "The Warwick Metadata Workshop: A Framework for the Deployment of Resource Description." D-Lib Magazine, July/August 1996.
[FU97] Furuta, R., Shipman, F., Marshall, C., Brenner, D., Hsieh, H., "Hypertext Paths and the World-Wide Web: Experiences with Walden's Paths," In Hypertext '97: the Eighth ACM Conference on Hypertext (Southampton, U.K., April 6-11 1997), pp. 167-176.
[HA96tr] Haynes, J., Wantz, L., Miller, M. "Courseware Description Language," IAI Technical Report (Rockville, MD), 1996.
[JO95] Joachims, T., Mitchell, T., Freitag, D., Armstrong R. "WebWatcher: Machine Learning and Hypertext," Fachgruppentreffen Maschinelles Lernen (Dortmund, Germany), August 1995.
[LI95] Lieberman, H. "Letizia: An Agent That Assists Web Browsing," International Joint Conference on Artificial Intelligence (Montreal), August 1995.
[MI98] Miller, M., Wantz, L. "COOL Links: Ride the Wave," In Proceedings of the Seventh International World Wide Web Conference (Brisbane, Australia), April 1998.
[SU98] Sutton, S. "Gateway to Educational Materials (GEM): metadata for networked information discovery and retrieval," In Proceedings of the Seventh International World Wide Web Conference (Brisbane, Australia), April 1998.
[TR88] Trigg, R.H. "Guided Tours and Tabletops: Tools for communicating in a hypertext environment." ACM Trans. Off. Inf. Syst. 6, 4 (Oct. 1988), 398-414
[WA98] Wantz, L., Miller, M. "Educational Metadata In Use," In Proceedings of the Seventh International World Wide Web Conference (Brisbane, Australia), April 1998.
[WA97] Wantz, L., Miller, M. "Toward User-Centric Navigation of the Web: COOL Links using SPI," In Proceedings of the Sixth International World Wide Web Conference (Santa Clara, CA), April 1997.
[WE97] Weibel, S. et. al. "The 4th Dublin Core Metadata Workshop Report." D-Lib Magazine, June 1997

URL References

COOL Links: http://www.i-a-i.com/services/cool/
Courseware Description Language (CDL): http://www.i-a-i.com/services/cdl/
Dublin Core: http://purl.oclc.org/metadata/dublin_core/
Gateway to Educational Materials (GEM): http://geminfo.org/
Instructional Management System (IMS) project: http://www.imsproject.org/
Letizia: http://lieber.www.media.mit.edu/people/lieber/Lieberary/Letizia/Letizia.html
NIST Learning Objects Metadata Group (LOMG): http://jetta.ncsl.nist.gov/metadata/
RDF http://www.w3.org/Metadata/
W3C: http://www.w3.org/
Walden's Paths: http://www.csdl.tamu.edu/walden/
Warwick Framework: http://www.ub2.lu.se/tk/warwick.html
Web Watcher: http://www.cs.cmu.edu/afs/cs.cmu.edu/project/theo-6/web-agent/www/project-home.html
XML: http://www.w3.org/XML/