Adaptive hypermedia - II

ocelotgiantAI and Robotics

Nov 7, 2013 (3 years and 8 months ago)

84 views

Adaptive hypermedia
-

II

-
Lost and found in hyperspace

by


Deepankar
-

09005060


Salman Khan
-

09005064


Haseeb
-

09005070

Contents

2


Introduction


Concept of hypermedia


Problems with hypermedia


Concept of adaptability


Adaptive hypermedia


concepts


Models in Adaptive hypermedia


Information retrieval


Conclusion


Introduction

3


Adaptive

hypermedia
, abbreviated as AH, is the next
-
generation of
hypermedia

applications.


Adaptive hypermedia is the answer to the
"lost in hyperspace"

syndrome, where the user has normally too many links to choose
from.

What is hypermedia?

4


Term coined by
Ted Nelson
in 1965.


Hypermedia is a style of building systems for organising, structuring and accessing
information around a
network

of multimedia nodes connected together by links


Conclin, 1987.


Analogy : Linear


Non
-
linear


Text
-

Hypertext , Link


Hyperlink : Media
-
Hypermedia.


“The WWW is fundamentally a distributed hypermedia application”


Taylor,
Medividovic, Dashofy(2010)













vs.


Problems in hypermedia

5


Hypermedia is largely
chaotic

in nature.


Problems are due to storage and retrieval of
large data
.


Disorientation

: When a user accesses the information, he may
easily disorient himself due to the
vast

information
.


Cognitive

overhead

: To stay oriented, the level of
involvement

of the user should be reasonably high, which leads in increased
cognitive overhead.


Both of them are
inter
-
related
.

Problems in hypermedia

6

Solution


Concept of Adaptability

7


As we saw earlier, disoriented navigation is root to many problems of
hypermedia.


Solution : provide users with an
orientation
-

Direct users to where
they should go.


Requires
dynamically

getting the information of what the user’s goal
is, inferred by his path.


This concept of
guiding

the user through the hyperspace


Concept
of
Adaptive

hypermedia.

What is adaptive hypermedia

8


“An attempt to analyze and personalize”


It is the concept of
integrating

hypermedia with user modeling.



Why adaptive hypermedia?

9


Can solve the problem of hypermedia systems which are used by
different
classes

of users, who can significantly differ in their
features

covered by the hypermedia system.


Adaptation can
prevent

the user from getting
lost

in hyperspace, by
providing navigation support by
limiting

browsing

space (i.e.
hiding non
-
relevant links).

What can be adapted?

10


The space for adaptation is quite limited. There are not so many
features which can be adapted.


Two types of adaptations :


Content
-
level

adaptation : Content of regular pages. Also called as
adaptive presentation. Solves problem of different classes of users
using same system.


Link
-
level

adaptation : Links from the regular pages. Also called as
adaptive navigation support. Solves problem of users getting lost in the
hyperspace.

Adaptive presentation

11


Adapt the
content

of the page to current knowledge level, goals and
other characteristics of the user.


A qualified user can get more and deeper information than a novice
user visiting same page at same time.

For Example…

12


For example, consider two people visiting some university
website.


If the visitor is a student, we show to him content like timetable,
recent events, rules and notices etc…


But, if the visitor is an outsider, we must show to him facilities
around the campus, different faculty profiles, gallery, awards and
recognitions received by college etc…

Adaptive navigation support

13


Helps users to find their paths in hyperspace by adapting the style of
link presentation
to the goals, knowledge etc...


Most popular techniques: Guidance, sorting, hiding and annotation.


Adaptive navigation support

14


Direct guidance
: To decide which is the
next

best node for the
user, depending on the parameters of the user.


Adaptive ordering
: To
sort

all links of a particular page, according
to user model, the closer to the top, the more relevant the link is.


Adaptive hiding
: To
hide

the links to the non
-
relevant pages,
thereby restricting the user to not go to irrelevant pages.


Adaptive annotation
:
Augmenting

links with some comments,
which gives users more details about the current state of the node
behind the annotated link.

Say,

15


Consider for example, two users open a virtual classroom website.


Say one of the users always works only on math, and another one only
on science.


Then on the webpage, the links can be modified such that only the
links relevant to the user’s area of interest are visible on the page.


Also, if we know user’s level of knowledge, and he wants to improve
it, we can show the links he must visit in some “
Recommended

section.

16

What is being modeled?

17


There are two broad classifications in the AH models :

1). Represent features of the user as an
individual



important to all
adaptive web systems.

2). Represent the user’s current
context

of the work


important to
mobile adaptive systems.



Knowledge based

18


User’s knowledge can both increase and decrease with time


User model should be
updated

accordingly.


User’s knowledge is generally estimated using
scalar model
, by
a single value on scale, either quantitative (0


5) or qualitative (
good, fair, bad etc..).


Users are then grouped based on the knowledge and are served
different versions of the same page.

The Domain model

19


It decomposes the body of knowledge into a set of domain knowledge
elements like concepts, knowledge items etc…


The domain knowledge or information is divided into elementary
fragments

called as
concepts
.


Independent concepts: set model or
vector

model. This is used for
less data, and is generally imprecise.


Semantic web applications: domain
ontologies

and domain
topic

maps.


Types of domain models

20


Vector maps
: Vector maps are generally used when the fragments of
knowledge are all
independent.


Topic maps
: A topic map represents a collection of topics, each of
which represents a concept.


They are related to each other by ‘
associations
’, which are
n
-
ary

combinations of topics.


Ontologies

: ontology


“Formal explicit specification of a
shared

conceptualization”


Example
-

Card : “poker” domain ontology interprets it to be playing
card, while “computer” domain interprets it to be punching card.

Overlay model of knowledge in AH

21


To represent user knowledge as a subset of the domain model, which
reflects the
expert’s

knowledge.


For
each fragment
of the domain’s knowledge, an overlay model
stores some estimation of the knowledge level of it’s.


Pure

model
: Boolean variable to indicate whether the user knows, or
doesn’t know the fragment.


In this case, user knowledge is represented as an exact subset, or

overlay
” of expert knowledge.

Example

22

Interest based

23


Important in information retrieval and filtering data.


Equal importance as user knowledge.


Two types of interest based models :

1).
Keyword
-
level

models : Models taking into account the
keywords used more frequently by the user.

2).
Concept
-
level

models : Models taking into account the concepts
related to user, like geographic location, native language etc…


Generally, concept
-
level models provide
more accurate
information
than keyword
-
level models.


Immediate

processing of data in
latter
.

Example

24


Keyword
-
level:
This model depends on keywords used by the user more
frequently.


Say, if the user always searches about novels and authors, then when he
searches for the word “The Associate”, he must get one of his major results to
be the novel by the same name, rather than it’s original meaning.


Concept
-
level:
This model depends on user concepts like location,
language etc…


Say, if an user searches for the word “bulls”, if he is an American, the result
must be “Chicago bulls”, while it should be animal bull otherwise.

Overlay model of user interests

25


Similar to overlay knowledge modeling, but differs in the structure
and size of the concepts.


Actually, domain models used for both overlay knowledge modeling
and overlay interest modeling are compatible.


Interest tracking

26


Three groups of models that do interest tracking.


First group :
Does the tracking using
vector

models. Uses a set of
independent concepts.


A demonstrated interest in some object is modeled as an increased interest in
related concepts.


Second group :
Uses
taxonomy

model, which is formed by classification
hierarchy of concepts.


An interest in machine learning indicates an interest in parent topic artificial
intelligence.


Third group :
Uses
ontology

model, which is formed by links among
interests, like related to, parent of, is similar to, etc..


An interest in Movies indicates an interest in TV etc…

Goals and tasks based

27


Goal : “What does the user actually want to achieve?”


It can change several times, even in one session.


The information flow from user is less, making it difficult.


One solution : Maintaining a user goal
catalogue
, and model the
user’s current goal as a
probabilistic

overlay

of the goal catalogue.


System maintains probability that the current goal is the goal of the
user.

Example

28

User’s background

29


Set of features
out of
the core
domain

of the system.


Range of background
-

User’s profession, job responsibilities, work
experience in related areas etc…


Example


Medical AH systems provide data in different ways for a
doctor and for a normal user.


Can be related to user’s knowledge.


Background is generally
static

while user’s knowledge of the domain
can be
dynamic
.

Individual traits

30


Features that define the user as an individual.


Cognitive

styles : It means an individually preferred and habitual
approach to information.


Example: impulsive/reflexive etc…


Learning

styles : The way people prefer to learn, as in through
pictures, videos etc…


There still are no proven systems for the application of learning styles
in adaptation.

Context of work

31


Takes into account several contexts such as user location, physical
environment, social context etc…


User
platform

: Adaptation to this is an important issue.


Hardware, software and bandwidth come under this.


User
location

: Modeling and usage of user’s location is slightly
different from other elements.


Coordinate
based, or zone
-
based user location.

Generalized overlay models

32


User features beyond knowledge and interests are not generally
modeled as a formal overlay of a domain model.


A generalized domain model is a
set of aspects
which can represent
domain concepts, domain tasks, goals etc…


Is a set of
aspect
-
value

pairs, where aspects are as above mentioned,
and values can be quantitative(value?) or qualitative( “yes” or “no”).


These are more
powerful

than the traditional models.

Uncertainty based user modeling

33


“The user failed to answer this question. So most probably, he doesn’t
know C++”


Uncertain

information.


“The user has been spending on
Facebook

for quite a long time”


Imprecise

observation.


Bayesian

networks : Developed by Judea Pearl in 1980’s. Based on
Bayes
’ theorem.
Causality

inspired probabilistic models.


Fuzzy

logic : Proposed by Professor
Lofti

Zadeh in 1965. Takes values
other

than 0 and 1. More closely related to imprecise observation.

Getting user data
-

some methods

34


Heuristic inference:
Based on weighting mechanisms that combine
the evidence about behavior to infer the features.


Depending on time spent on each webpage, user interest can be
inferred.


Activation/Inhibition networks:
Where domain models are
used, user feedback is taken as direct interest in domain concepts, and
propagated over the links to related concepts.

Some more!

35


Collaborative filtering:
User model is updated by the items
selected by the user, directly.


Content
-
based filtering:
User model evolves on the basis of the
features of the selected items


“Bag of words” representation


Rocchio

Algorithm.


Semantic reasoning
: Based on the ontological representation of the
domain, which is used to reason about the user’s features, interests,
etc… by traversing semantic relations among concepts.

Rocchio

Algorithm

36


Based on method of
relevance feedback
found in information
retrieval systems.


From
SMART (
Self
-
Monitoring, Analysis and Reporting
Technology
)

information retrieval system.


Explanation

37


Constants a, b and c have a role in shaping the modified vector in a
direction closer or farther away from original query, related and non
-
related documents.


In general,

a = 1

b = 0.8

c = 0.1.


Metrics of Information retrieval

38

F
-
Score (F
-
Measure)

is something that considers both the above factors. Its close to
their harmonic mean.

Conclusion

39

Resources

40


www.cs.brown.edu/memex/ACM_HypertextTestbed/papers/25.html


en.wikipedia.org/wiki/Hypermedia


en.wikipedia.org/wiki/Adaptive_hypermedia


www.elearnspace.org/blog/2003/07/17/adaptive
-
hypermedia/


www.sics.se/~kia/papers/edinfo.html


tihane.wordpress.com/2007/01/08/patterns
-
of
-
interaction
-
hypermedia
-
and
-
web20/



journals.tdl.org/
jodi
/article/
viewarticle
/6/6


http://www.stonetemple.com/how
-
google
-
does
-
personalization
-
with
-
jack
-
menzel/



http://msdn.microsoft.com/en
-
us/library/aa480048.aspx



Resources

41


aetos.it.teithe.gr/~cs1msa/hyp0.html


dl.acm.org/
citation.cfm?id
=948459


www.rc.au.net/papers/www
-
0595/wwwhype2.html


www.amundsen.com/blog/archives/1069


roberson.wikispaces.com/
Advantages+of+Hypermedia


aima.eecs.berkeley.edu/slides
-
ppt
/m14
-
bayesian
.ppt


www.cse.unr.edu/~bebis/CS365/StudentPresentations/
FuzzyLogic
.ppt

References

42


User models for Adaptive hypermedia and adaptive educational systems


Peter
Brusilovsky

and Eva
Millan
, University of
Maliga,USA
.


Adaptive hypermedia : An attempt to analyze and generalize


Peter
Brusilovsky
,
ICSTI Russia.


Personalization in cultural heritage : the road travelled and the one ahead


Liliana

Ardissono
,
Tsini

Kuflik
, Daniela
Petrelli
, Springer 2011.


Modal navigation for hypermedia application


Franca
Garzotto
, Luca
Mainetti
,
Paolo
Paolini
, University of Lecce, Italy.


From Adaptive hypermedia to the adaptive web


Peter
Brusilovsky
, Mark T.
Maybury
, Communications of the ACM, may 2002.


Methods and techniques of adaptive hypermedia


Peter
Brusilovsky
, 1996
Kluwer
.


Adaptive hypermedia


Peter
Brusilovasky
, 2001
Kluwer
.


43