Using Adaptive Hypermedia for Web-based Education

1
Using Adaptive Hypermedia for Web-based Education

• Paul De Bra
• Eindhoven Univ. of Technology
• Peter Brusilovsky
• Carnegie Mellon University

2
Topics

• Why adaptive hypermedia systems (AHS)?
• Adaptive hypermedia design
• Implementation issues for AH on the Web
• The AHA system
• Authoring for AHA
• AHA demo
• Authoring for Interbook
• Interbook demo

3
Why Adaptive Hypermedia?

• Adaptive hypermedia has two purposes
• prevent users from getting lost by means of
  individualized navigation and orientation
  support.
• adapt the information on the pages to each user,
  depending on the order in which pages are read.
• Adaptive hypermedia systems therefore offer
• adaptive maps (fish-eye views) of the overall
  link structure (and indication where the user
  is).
• manipulation of the links (annotation, hiding,
  ...)
• conditional inclusion (or emphasis or
  modification) of parts of a pages contents.

4
Adaptive hypermedia design

• We follow the AHAM reference model
• domain model describes the subject domain at the
  conceptual level.
• user model describes the users knowledge,
  preferences, background, experience, etc.
• teaching model describes how adaptation should
  be done, depending on the domain model and the
  user model.

5
Domain model

• It describes the subject domain in terms of
  concepts.
• atomic concept fragment of information, that
  cannot be changed by the AHS.
• page (concept) a collection of atomic concepts
  (or fragments), structured in some way (e.g. a
  sequence) fragments are conditionally included
  in a presentation.
• composite concept a collection of other
  composite concepts and/or pages, structured in
  some way.

6
Structure of a composite concept
WWW
HTML
HTTP
(set of pages)
what
tags
specials
request
response
(sequence of fragments)
(sequence of fragments)
7
Constraints for concepts

• A concept may not contain itself (directly or
  indirectly).
• A composite concept may only contain other
  composite concepts or pages, not atomic concepts.
• A page may contain only fragments.
• A concept may occur in different composites.
• A fragment (atomic concept) may appear on
  different pages.

8
Concept relationships

• Concepts relate to each other in different ways
• link relationship there may be a hypertext link
  from a concept to another concept.
• prerequisite relationship the user should read
  (or learn about) some concept(s) before starting
  on some other concept(s).
• inhibitor relationship once a user has read (or
  learnt about) some concept(s) some other
  concept(s) become irrelevant.
• there may be many more relationship types...

9
Constraints for concept relationships

• For each type of concept relationship there are
• different constraints
• A link must have at least one source and one
  destination.
• A concept must never be a prerequisite for
  itself, either directly or indirectly.
• A concept must never inhibit itself, either
  directly or indirectly.
• There may be many more constraints for these or
  other relationship types.

10
Notes about domain model

• Composite concepts must have a structure when
  the concept is the destination of a link the
  structure is used to determine a sub-concept as
  the actual destination.
• Pages must also have a structure when presenting
  a page the structure determines in which order to
  present the fragments.
• Concept relationships do not indicate how they
  influence the adaptation the teaching model
  expresses the adaptation.

11
User Modeling

• An adaptive hypermedia system is usually also
  adaptable some aspects of the user model can be
  explicitly set by the user (through a form or
  questionnaire).
• The adaptive features normally concentrate on the
  knowledge about concepts. An AHS monitors the
  users actions
• requests for pages
• reading time per page
• results for (multiple-choice) tests

12
User Modeling (cont.)

• A user model is more or less a table
• for each user and each concept a number of
• attribute values are stored. Example

...
name
knowledge
read
ready-to-read
WWW
learned
false
false
HTML
well-learned
true
true
HTTP
not-known
false
true
...
...
13
User modeling (cont.)

• There are three common ways to represent the
  users knowledge about a concept
• Boolean model a concept is known or not-known.
• Discrete model there are a few values, e.g.
  well-known, well-learned, learned, not-known.
• Continuous model there is a range of values,
  e.g. the interval 0..1, or a percentage (values
  between 0 and 100) as an approximation.
• Beware of too much detail the source of
  user-model information is not reliable.

14
User modeling (cont.)

• The representation of knowledge influences the
  granularity of the domain model
• With the Boolean model every page is associated
  with a domain model concept there is
  little use for composite concepts.
• With the discrete model there is limited use for
  composite concepts.
• With the continuous model pages may contribute a
  different amount of knowledge to a composite
  concept. The reading time (per page) may be
  taken into account as well.

15
User modeling (cont.)

• other attributes in a user model
• read this stores information about what was read
  by the user. For a composite concept nothing is
  read. For a page read may be
• Boolean read or not read.
• access times as in a log file.
• reading time actually display time.
• ready-to-read stores information about whether
  the user is ready to learn about this concept.
  May be Boolean, discrete or continuous (0..1).

16
The teaching model

• an AHS performs adaptation based on the domain
  model and the user model. It follows certain
  rules, e.g.
• When there is a link from C1 to C2 then there
  must be a link anchor in C1 for a link to C2.
• When C1 is a prerequisite for C2 then links to C2
  are not desirable until enough knowledge about C1
  is acquired.
• When a page is read that is desirable more
  knowledge is acquired than when reading an
  undesirable page.

17
The teaching model (cont.)

• there are two kinds of rules
• Some rules are executed before the presentation
  of the next page. They can be used for
• determining which page to display.
• selecting and ordering fragments to display.
• selecting how to present anchors to which
  concepts.
• Some rules are executed after the presentation of
  the next page. They are used for
• updating the user model.

18
The teaching model (cont.)

• Selecting a link destination involves
• When the destination is a composite concept one
  or more sub-concepts must be chosen to be the
  actual destination to presented.
• This process is repeated until the actual
  destination is a page.
• When the (actual) destination is a page the
  content of the page is constructed by selecting
  fragments and possibly sorting them (if the page
  structure is not already a sequence).

19
The teaching model (cont.)

• Constructing a page is done by
• For each fragment it is decided whether or not it
  is desirable. This is done by checking
  prerequisite, inhibitor and possibly other
  requirements.
• The selected fragments are sorted. If the page
  has a sequence structure then the order of the
  fragments is fixed. But other structures are
  possible which require the order of fragments to
  be computed in some way.

20
The teaching model (cont.)

• Whether or not a concept is desirable is
  determined by rules like
• if C1.knowledge lt 0.8 and C1 is prerequisite for
  C2 then C2 is not desirable.
• if C1.knowledge gt 0.8 and C1 is inhibitor for C2
  then C2 is not desirable.
• There are propagating (transitive) effects
• if C1 is prerequisite for C2 and C2 is
  prerequisite for C3 then C1 is prerequisite for
  C3. So when C1 is not known then C3 is not
  desirable.

21
The teaching model (cont.)

• There are four ways to deal with anchors for
  links leading to undesirable concepts
• hiding the anchor is hidden by making it
  indistinguishable from normal content (text).
• annotation the anchor is presented differently
  from anchors of links to desirable concepts. (The
  color may be different for instance.)
• removal the anchor is removed from the content
  of the page. (May disrupt the content.)
• disabling the anchor does not activate a link,
  but it may still look like a link anchor.

22
The teaching model (cont.)

• Link manipulation (annotation) can be expressed
  by rules, e.g.
• if A1 is anchor of link from C1 to C2
• and C2 is desirable
  
  then if C2 was not read then A1.color green
  
• else A1.color yellow
• else A1.color red
• Link hiding is like link annotation, but the
  undesired color is black (or whatever the color
  of normal text is).

23
The teaching model (cont.)

• The rules for updating the user model are
  executed after generating a page.
  Example (for concept C of page P)
• if P is just read
• then if C is desirable then
• C.knowledge 0.8
• else C.knowledge 0.5
• else C.knowledge remains unchanged
• There may be other rules for changing / storing
  read, desirable and other values.

24
The teaching model (cont.)

• Rules may be built into the AHS or written by the
  author.
• Rules may be generic for all relationships of a
  certain type, or specific for some specific
  concepts.
  Specific rules may be
  intended to create exceptions to generic rules.

25
Implementing AH on WWW

• There are three well-known and general
  server-side technologies
• CGI-scripts a process is executed for each
  HTTP-request it generates one HTML page. There
  is process and communication overhead.
• Fast-CGI a script-server process handles each
  HTTP-request it generates the HTML pages. There
  is only communication overhead.
• Servlets some Java code is added to the
  Web-server it generates the HTML pages.
• Programming for these technologies is similar.

26
Implementing AH on WWW (cont.)

• An adaptive engine works as follows
• 1. First the stored user model is retrieved from
  disk (from a file or database).
• 2. Other attributes used in rules are
  initialized.
• 3. The requested concept is resolved to a page.
• 4. Rules are executed to select the fragments to
  be presented (and possibly to sort them).
• 5. Rules are executed to manipulate the link
  anchors (like annotate or hide them).
• 6. Rules are executed to update the user model.
• 7. The user model is saved (in file or database).

27
Dealing with adaptive content

• A generated HTML page (containing the information
  and the links) may be different each time it is
  requested.
• The browser must request the page (from the
  server) each time the user requests it.
• Intermediate proxies must not cache the page.
• One solution is to use a different URL each time
  the page is requested. This makes it impossible
  to bookmark a page.
• A better solution is to force browser and
  proxies to not do caching.

28
Link manipulation in HTML

• Different links on the same page need to be
  displayed in a different way (to indicate whether
  the destination is desirable.
• Interbook annotates anchors with colored dots
  and arrows .
• AHA uses colored link anchors. It uses cascading
  style sheets (CSS) to color links depending on a
  link class.
• Links can be hidden by removing the anchor tag
  ltA...gt.

29
Working with Frames Interbook
table of content
naviga-tion help
pre-requisite concepts
infor-mation page
generated concepts
30
Working with Frames Interbook

• Interbook uses a fixed layout with five frames
  four are generated by Interbook and one is
  (mostly) written by the author. The automatic
  generation of auxiliary frames makes authoring
  easy.
• Each time a request is made from one frame four
  frames must be updated, through four HTTP (almost
  simultaneous) requests to the server. This puts a
  heavy load on the server.

31
Working with Frames AHA

• AHA does not enforce any frames structure.
• AHA does not offer automatic generation of frames
  (structure or content).
• In each frame one can display a generated header
  and footer, or omit them.
• JavaScript (or VBscript) code is used to
  synchronize updates to frames. Updating different
  frames puts a heavy load on the server.

32
The AHA system

• Each concept corresponds to a page.
• For each page there is a precondition which is
  expressed in terms of concepts.
• Each page generates knowledge about one or more
  concepts.
• Fragments of content (text) are included based
  upon preconditions per fragment.
• Conditional links are hidden when the
  precondition for their destination is not
  fulfilled. This can be changed to annotation.

33
The AHA system (cont.)

• Content is authored as HTML files with special
  comments
• lt!-- requires readme and intro but not www --gt
• lt!-- generates html --gt
• lt!-- header --gt
• This page becomes desirable when the readme
• and intro have been read and www has not.
• The page generates knowledge about html.
• The text appears between a header and a footer.
• lt!-- footer --gt

34
The AHA system (cont.)

• Text fragments are conditionally included
• lt!-- if www and not (html or http) --gt
• This text sometimes appears
• lt!-- elseif www and not (html and http) --gt
• and sometimes this
• lt!-- else --gt
• and otherwise this
• lt!-- endif --gt
• These if statements can be nested.

35
The AHA system (cont.)

• AHA generates a CSS style sheet for links
• ltstyle typetext/cssgt
• A.good color 0000ff
• A.bad color 202020
• A.neutral color 7c007c
• lt/stylegt
• The HTML contains conditional links
• lta classconditional href...gtanchor textlt/agt
• This is turned into (in case of a desired link)
• lta classgood href...gtanchor textlt/agt

36
AHA components

• The AHA system document consists of
• a Makefile to generate concept and dependency
  lists.
• a welcome file with registration form.
• a header file with references to progress report
  and adaptable features.
• a footer file with reference to the Stop
  applet.
• a html directory containing the pages.
• an icons directory containing standard images.
• an images directory containing other images.
• an applets directory containing the Stop
  applet.
• a reference to the servlet doing the adaptation.

37
AHA components (cont.)

• The welcome file must contain (something like)
• Welcome to this AHA document.ltbrgt
• ltform methodpost actionservlet-namegt
• name ltinput typetext namename size40gtltbrgt
• email ltinput typetext nameemail
  size40gtltbrgt
• id ltinput typetext nameid size10gtltbrgt
• password ltinput typepassword namepasswd
  size10gtltbrgt
• ltinput typesubmit valuego!gt
• lt/formgt

38
AHA components (cont.)

• The header file looks like
• lta href...gtltimg src/.../icons/smalltue.gifgtlt
  /agt
• lta href/.../cgi/get/ID/config/generategt
• ltimg src/2M350/icons/config.gifgtlt/agtltbrgt
• AHA adds references to a progress report
• ltbgtJohn Doe has read 47/159 pages.lt/bgt
• (lta href__done__gt these readlt/agt,
• lta href__todo__gtthese still to dolt/agt.)

39
AHA components (cont.)

• The footer file can contain anything.
• AHA adds a Stop applet
• ltapplet codeStop.class codebase/.../applets
  width2 height2gt
• ltparam namestopurl value/.../stop/pagenamegt
• ltparam namehost value127.0.0.1gt
• ltparam nameport value8001gt
• lt/appletgt
• The Stop applet sends a request to the server
  when the user leaves the page.

40
AHA components (cont.)
Example of what a header may look like
Example of what a footer may look like
Note that the Stop applet does not show.
41
AHA components (cont.)

• How the Stop applet works
• A Java applet is initialized when its page is
  loaded. This is done through the init method.
• The applet is started each time its page is
  displayed. This is done through the start method.
• The applet is stopped each time its page is
  replaced by another one. The applet can still do
  something before really stopping the stop method
  is called.
• In the AHA Stop applet the stop method sends a
  request to the server, with the name of the page
  the user has stopped reading.

42
AHA components (cont.)

• The AHA servlet uses the following files to
  generate file x.html for user doe
• The set of existing concepts is read from a file
  knowlist.
• The set of concept relationships (dependencies)
  is read from a file doclist.
• The users knowledge is read from doe_model.
• The users logfile (with all access and stop
  times) is read from doe_log.
• The html file x.html, icons, images, etc. are
  read and turned into an html file for
  presentation.

43
Authoring for AHA

• Authoring requires the following steps
• Creating (copying and modifying) a welcome page,
  header, footer, and possibly an index page with
  the frameset.
• Writing the content, using any html editor or
  generator. The editor must support html comments
  and link classes.
• Removing possible html headers, and adding the
  lt!-- requires --gt lt!-- generates --gt,
  lt!-- header --gt and lt!-- footer --gt tags where
  appropriate.

44
Configuring the Servlet

• The servlet needs six parameters
• HTDOCS path of the documentroot of the
  Web-server.
• LOGS path of the log directory of the
  Web-server.
• TITLE the title for the whole adaptive
  hyperdocument. It is added to every page.
• COURSE the name of the directory (relative to
  documentroot) which holds the adaptive document.
• BACKGROUND the name of a background image.
• LANGUAGE for progress report (default English).

45
Setting up the browser

• In order to use AHA the browser must be
  configured as follows
• The browser must accept text colors and link
  colors specified in the html pages. (Some
  browsers let you disable this.)
• The browser must not underline links. (AHA hides
  links by making the text almost black, but
  underlining links would make them visible.)
• You must ensure that the browser requests a page
  each time it is needed. (AHA generates expired
  pages, but a browser may sometimes ignore this.)

46
Example page from course 2L690
47
Example page from course 2M350