Transcript CSA4080: Adaptive Hypertext Systems II - Search

CSA4080:
Adaptive Hypertext Systems II
Topic 2: User-Adaptive Systems
Dr. Christopher Staff
Department of Computer Science & AI
University of Malta
CSA4080: Topic 2
© 2004- Chris Staff
1 of 20
[email protected]
University of Malta
Aims and Objectives
• To consider where AHSs fit in the realm of
User-Adaptive Systems
• To describe other systems that adapt to the
user
CSA4080: Topic 2
© 2004- Chris Staff
2 of 20
[email protected]
University of Malta
User-Adaptive Systems
• Systems that adapt to their environment are
called Adaptive Systems
– e.g., Artificial Life
• Systems that adapt to their users are called
User-Adaptive Systems
– e.g., Adaptive User Interfaces, Recommender
Systems, Reconnaissance Agents, Adaptive
Information Retrieval, User modelling,
personal assistants, personalisation, information
filtering, ambient intelligence...
CSA4080: Topic 2
© 2004- Chris Staff
3 of 20
[email protected]
University of Malta
User-adaptive systems: Functions
•
•
•
•
•
•
•
•
Help user to find information
Recommend products to user
Tailor information presentation to user
Help user to learn about a topic
Help user to use a system
Adapt an interface to user
Perform routine tasks on behalf of user
Support collaboration between user and other
persons
ijcai01-tutorial-jameson.pdf
CSA4080: Topic 2
© 2004- Chris Staff
4 of 20
[email protected]
University of Malta
Typical Properties:
Adaptive User Interfaces
• As Graphical User Interfaces become more
complex, users need more help with the
interface
• Adaptive user interfaces learn a user model
by tracing the interactions with the interface
• They learn to improve their ability to
interact with a user
adapt.um99.pdf
CSA4080: Topic 2
© 2004- Chris Staff
5 of 20
[email protected]
University of Malta
Typical Properties:
Adaptive User Interfaces
• Examples of AUIs are:
– Recommendation systems, Syskill & Webert
– Personalisation systems, Calendar Apprentice
– Content-based filtering, NewsWeeder
CSA4080: Topic 2
© 2004- Chris Staff
6 of 20
[email protected]
University of Malta
Typical Properties:
Adaptive User Interfaces
• AUIs concentrate on how the user model is
learnt
• So concentrates on the user interaction, and
hence the interface between user and
machine
CSA4080: Topic 2
© 2004- Chris Staff
7 of 20
[email protected]
University of Malta
Typical Properties:
Reconnaissance Agents
• E.g., Letizia, PowerScout (Why-SurfAlone.pdf)
• Reconnaissance agents: “programs that look
ahead in the user’s browsing activities and
act as an advance scout to save the user
needless searching and recommend the best
paths to follow.” Why-Surf-Alone.pdf
CSA4080: Topic 2
© 2004- Chris Staff
8 of 20
[email protected]
University of Malta
Typical Properties:
Reconnaissance Agents
• Provide local and/or global guidance
• Typically, less user involvement in
identifying interest is better
– E.g., search engine usually requires active role
– Reconnaissance agent observes user to learn
model
CSA4080: Topic 2
© 2004- Chris Staff
9 of 20
[email protected]
University of Malta
Typical Properties:
Adaptive Information Retrieval
• Can bridge vocabulary ‘gap’ by learning
associations between user and document
vocabulary
• Can rephrase user query based on user
interactions with docs in results set
• Can provide ‘context’ for user terms to
disambiguate terms
CSA4080: Topic 2
© 2004- Chris Staff
10 of 20
[email protected]
University of Malta
Typical Properties:
Recommender Systems
• E.g., IMDB, Amazon, ...
• Two main types, but with same aim
– Collaborative vs. content-based
– Aim to make recommendation to individual,
based on past events
recommender 0329_050103.pdf
CSA4080: Topic 2
© 2004- Chris Staff
11 of 20
[email protected]
University of Malta
Typical Properties:
Recommender Systems
• Collaborative
– System will make recommendation based on
what other similar user have done
– User similarity: demographic vs. interaction
history
– Uses ratings
– If X & Y gave similar ratings for A, C, D, then
recommend F to Y if X liked F
CSA4080: Topic 2
© 2004- Chris Staff
12 of 20
[email protected]
University of Malta
Typical Properties:
Recommender Systems
• Content-based
– Also uses ratings, but we recommend F to Y, if
Y gave high rating to items in , where  is set
of objects similar to F
• Collaborative recommender systems suffer
if item is unrated
• Content-based systems suffer is user has no
history
CSA4080: Topic 2
© 2004- Chris Staff
13 of 20
[email protected]
University of Malta
Typical Properties:
Personal Assistants
•
•
•
•
User delegates work to the computer
Find and filter information
Customise views of information
"They enable users to center their interactions at
the content level (semantics), partially removing
syntactic difficulties. They also enable users to
index (contextualize) content to specific situations
that they understand better (pragmatism)" Boy,
Guy A. (1997) Software Agents for Cooperative
Learning. In Software Agents , MIT Press (1997)
CSA4080: Topic 2
© 2004- Chris Staff
14 of 20
[email protected]
University of Malta
Typical Properties:
Personalisation
• Changing view/interface/content to needs
and requirements of user
• Can apply to anything
CSA4080: Topic 2
© 2004- Chris Staff
15 of 20
[email protected]
University of Malta
Typical Properties:
Information Filtering
• Inverse function of information retrieval
• Constant stream of changing information
(e.g., a news wire) where each item needs to
be sent to an interested user
– Wrong item to user, user becomes overloaded
– Item not sent to user, user misses information
CSA4080: Topic 2
© 2004- Chris Staff
16 of 20
[email protected]
University of Malta
Typical Properties:
Ambient Intelligence
• Devices that adapt to changes in their and
their user’s environment
CSA4080: Topic 2
© 2004- Chris Staff
17 of 20
[email protected]
University of Malta
Conclusions
• So, there is lots of overlap in the different
fields of user-adaptive systems
– We haven’t talked about adaptive hypertext
systems, intelligent tutoring systems, on-line
help systems, ...
CSA4080: Topic 2
© 2004- Chris Staff
18 of 20
[email protected]
University of Malta
Conclusions
• What properties of the user should be modelled?
• What input data about the user should be
obtained?
• What techniques should be employed to make
inferences about the user?
• What functions are to be served by the adaptation?
• How should decisions about appropriate adaptive
system behaviour be made?
• What empirical studies should be conducted?
ijcai01-tutorial-jameson
CSA4080: Topic 2
© 2004- Chris Staff
19 of 20
[email protected]
University of Malta
For next lecture
• Read http://ted.hyperland.com/buyin.txt
CSA4080: Topic 2
© 2004- Chris Staff
20 of 20
[email protected]
University of Malta