Adaptive%20Systems

1
Adaptive Systems

• Paul De Bra

2
Topics

• Motivation
• Examples
• Adaptation methods
• Reference model
• Current research problems
• Conclusions

3
Motivation

• Old paradigm
• users adapt to the systems they use
• they accept non-intuitive user-interfaces
• they accept interfaces that never learn.
• New paradigm
• users expect configurable systems
• users expect systems to improve over time.

4
Examples

• Indicators for battery status
• (electric shaver, laptop computer, cellphone).
• TV / VCR
• suggests / records your favorite programs
• mutes when the phone rings, etc.
• Car
• seat position, radio volume, climate control.

5
Examples (cont.)

• Electronic Program Guide
• only provides info on available TV channels
• recommends programs you like (and omits programs
  you dont like)
• shows the kind of information you want (story,
  actors, director, text, photograph or video)
• adapts presentation to your device (computer with
  large screen, palmtop, WAP phone).

6
Examples (cont.)

• Real-estate database
• shows property in the area of interest
• learns which property you are likely to look into
  (based on price, kind, neighborhood...)
• knows which property you saw before
• shows only those attributes you want to see
• adapts presentation to your device / network.

7
Types of Adaptive Systems

• Embedded Systems
• User-Interface (placement of controls changes)
• Dialog (steps are inserted/deleted/reordered).
• Information Systems
• Query adaptation (select wanted items)
• Presentation adaptation (attribute selection and
  layout).

8
Adaptive Hypermedia

• Whats wrong with plain hypermedia?

A
B
9
Adaptive Hypermedia (cont.)

• How does adaptive hypermedia help?
• Provide additional explanations
• Provide prerequisite explanations for users who
  have a knowledge deficiency.
• Provide comparitive explanations.
• Provide explanation variants.
• Sorting present most wanted items first.

10
Adaptive Hypermedia (cont.)

• How does adaptive hypermedia help?
• Global guidance suggest navigation paths
• Local guidance suggest a page (or small set of
  pages) to be read next
• Global orientation support fish-eye view
• Local orientation support show where the user
  can go next.

11
Adaptive Hypermedia (cont.)

• Is creating adaptive hypermedia more work?
• Creating a sloppy website takes little effort
  (but users get lost, dont understand things,
  etc.)
• Creating an excellent static website with the
  right information and structure for everyone
  requires a lot of effort (and it still fails)
• Creating an adaptive site is somewhere in
  between...

12
Adaptive Hypermedia Architecture Model

• Domain model describes content and structure of
  the information
• User model describes how the user relates to the
  application
• Adaptation model describes how the information
  is presented to each user

13
AHAM Domain Model

• Atomic units of information fragments
• Fragments are combined into pages
  a page constructor describes how a page is
  built from conditionally included fragments
• A set of pages corresponds to a concept
  a set of concepts may represent a higher
  level composite concept.

14
AHAM Domain Model

• Concepts (and pages) are connected through
  concept relationships
• Links are a common type a link to a composite
  concept must be resolved to a destination page
  through a page selector
• Other relationship types can be defined, e.g.
  prerequisites, inhibitors, etc.

15
AHAM User Model

• There is a model for each user
  (we do not consider group models)
• For each concept there are a (possibly different)
  set of attribute/value pairs
• knowledge how well does the user know this?
• read has the user read something about this?
• ready-to-read may the user read about this?

16
AHAM Adaptation Model

• Rules that work like database triggers
• on event (e.g. user accesses a page)
• if condition is true (e.g. page was ready to be
• read, or desired)
• then execute an action (augment the
• knowledge of this page)

17
AHAM Adaptive Engine

• How are these rules executed ?
• The system searches for an active rule
• It executes one or more active rules in parallel
  (set semantics or instance semantics)
• The system then again searches for an active
  rule, etc. until no more active rules are
  available.

18
AHAM Adaptive Engine

• Termination of rule execution
• Does the rule execution always terminate?
  e.g. integer monotonic updates with upper
  bound
• Is termination decidable?
  e.g. only finite attribute
  domains
• If decidable, is detecting (non)termination
  tractable?

19
AHAM Adaptive Engine

• Confluence of rule execution
• When more than one rule is active, does the final
  result of the rule execution depend on which rule
  is executed first?
• When executing rules in parallel, can the
  resulting updates be conflicting?
• Does assigning a priority to rules help?

20
Termination in general

• Study the activation or triggering graph
• nodes are rule instances
• edges connect rules when the action of one rule
  triggers another rule and the condition of that
  other rule may be satisfied
• when this graph is acyclic termination is
  guaranteed (but the converse is not true).

21
Confluence in general

• The result of an event may depend on the order
  of execution of triggered rule
• when two rules can be executed, the action of the
  first rule may make the condition of the second
  rule become false.
• there is some (active database) theory for
  deciding whether all rule instances commute.

22
Applications at the TU/e

• 2L690 Hypermedia Structures and Systems
  (adaptive such that printing is not advisable)
• 2R350 Graphical User-Interfaces
  (one source for full text and viewgraphs)
• IShype Guide for internships and masters thesis
  work
• Corporate TU/e website is next!

23
Finally

• Several adaptive hypermedia applications exist
  today. (Web-based or non-Web-based)
• Adaptive hypermedia is no longer a niche novelty.
  (August 2000, AH2000 conference, Trento, Italy.
  LNCS 1892)
• Some theoretical and development work is still
  needed to create general-purpose AHS.