1. CSA3212: User Adaptive Systems
Lecture 8: Case Studies
Dr. Christopher Staff
Department of Computer Science & AI
University of Malta

2. Aims and Objectives
Adaptive navigation in Letizia, Personal WebWatcher, WebWatcher, and HyperContext
Adaptive Presentation in MetaDoc

3. Aims and Objectives
We will look at three different approaches to adaptive Hypertext
Adaptive navigation using link recommendation
Personal WebWatcher
Adaptive presentation using stretchtext
MetaDoc
Context-based adaptive navigation
HyperContext

4. Adaptive Navigation
Adaptive Navigation-local reconnaissance is highly related to link annotation
E.g., Letizia, WebWatcher, Personal WebWatcher, HyperContext

5. Adaptive Navigation
Differences in ITS and generic approaches to adaptive navigation
ITS aim is to transfer knowledge efficiently by guiding through a learning space
Learned, ready to be learned, not ready to be learned
Generic aim is to guide user through document space to relevant information (that is ideally also at the level of simplicity required by user!)
Relevant, not relevant (what about “related to long-term interest X?”)

6. Adaptive Navigation Letizia
Predicts a user’s interest as the user browses and suggests links to relevant document in the vicinity of the user’s current location
User tends to traverse Web graph “downwards”, but relevant information may lie sideways
Observes user behaviour to determine user interests (eg, “skipping” links, bookmarking...)
Makes recommendations based on “persistence of interest”
Persistence of interest: just because you’ve stopped searching for something, doesn’t mean you’re no longer interested in it...
lieberman95letizia.pdf

7. Adaptive Navigation WebWatcher
Guides users through a web site based on interaction with past users
Users express a query and are guided to relevant documents
Associates what users are interested in with documents that they mark as relevant
Marks up links with terms used by users, and terms that occur in “downstream” documents
webwatcher.ijcai97.pdf

8. Personal WebWatcher
Personal WebWatcher recommends documents to a user based on an analysis of the documents that the user has browsed
References:
Mladenic, D. (1996), Personal WebWatcher: design and implementation. Available on-line at
Mladenic, D. (1999), Machine learning used by Personal WebWatcher. Available on-line at
Additional information about Personal WebWatcher can be found at

9. Personal WebWatcher
PWW observes users of the WWW and suggests pages that they may be interested in
PWW learns the individual interests of its users from the Web pages that the users visit
The learned user model is then used to suggest new HTML pages to the user

10. Personal WebWatcher Architecture a Web proxy server a learner
The proxy saves URLs of visited documents to disk
a learner
The learner uses
them to generate
a model of user interests
When a user visits a Web page, PWW’s proxy server also analyses out-links
Recommends those similar to user model

11. Learning the user model
Operates in batch mode
Revisits all documents visited by user and those lying one link away
Visited documents are +ive examples of user interests
Non-visited are -ive examples

12. Personal WebWatcher
Model used to predict if a page is likely to be relevant (+ive) or not (-ive)
Predictor looks one step ahead from document requested by user
Links in requested document are marked up

13. HyperContext
HyperContext assumes that the scope of relevance within a document is dependent on its context
Remember that information is data in context…
… knowledge is information used in the correct context

14. HyperContext
HyperContext also assumes that a link is evidence that the destination document is relevant to the parent (in some way)
Is all of a document relevant in its entirety to all of its parents?
HyperContext says not.
Can semi-automatically determine which regions in the child are relevant to the parent

15. HyperContext Context is used in two ways
To create interpretations of documents in context
Interpretation = relevant terms from parent added to child, and remove non-relevant terms from child
To construct a short-term model of user interests as a user browses through hyperspace
Pick up relevant terms from the interpretations that are visited and “add” them to user model

16. HyperContext
Interpretations, as well as original documents, are indexed
Query can be automatically extracted from user model and submitted to IR system
User can be guided to relevant information (link recommendation), or shown “See Also” references

17. HyperContext
Uses Information Retrieval-in-Context to guide users to information in hyperspace (up to 7 link traversals away)
Once user has navigated to a location which probably contains information, can submit query to search “context sphere”
With Adaptive Information Discovery, system generates query on behalf of user
HCTCh5.pdf

18. Adaptive Presentation
Approaches are generally intended to make the content more understandable to the user
automatically including glossary explanations of terms unknown to the user
removing extraneous information, or information known to the user
showing information in format preferred by user...
Used mainly in ITS

19. MetaDoc Adaptive presentation of text
Documentation reading system that has hypertext capabilities
Reference:
Boyle, C., and Encarnacion, A.O., 1994, “Metadoc: An Adaptive Hypertext Reading System”, in Brusilovsky, et. al. (eds), Adaptive Hypertext and Hypermedia, 71-89, 1998, Netherlands:Kluwer Academic Publishers.

20. MetaDoc
Goal:
“A hypertext document that automatically adapts to the ability level of the reader”
No need for reader to “skip” text, or to look elsewhere for further information

21. MetaDoc Mechanism: Coined by Ted Nelson, 1971
Stretchtext
Coined by Ted Nelson, 1971
Transitions from one level to the next need to be smooth (HCI)
User model used to determine ability level of user

22. MetaDoc User Model: Concept Level:
Stereotypes: Novice, beginner, intermediate, expert
Concept Level:
Concept levels are associated with stereotypes
If user level is lower than the level required to understand the concept, the text is stretched to explain it
Conversely, more detail is provided to the expert reader