NOTES
ABOUT PERSONALIZATION & CUSTOMIZATION
Sources:
Haym
Hirsh, Chumki Basu, and Brian D. Davison "Learning to Personalize"
ACM special issue on Personalization, 43 (8), Aug
2000
Barry
Smith & Paul Cotter: "A Personalized Television Listings
Service", ACM special issue on Personalization, 43
(8), Aug 2000
Udi
Manber, Ash Patel, & John Robison: "Experience with Personalization
on Yahoo!" ACM special issue on Personalization,
43 (8), Aug 2000
Differences
about Personalization and Customization
Although
both terms have been used without distinction, the IA and UI
communities have been trying to develop the two concepts and
clarifing the differences aand impact of those two methods in
providing to users a better individual experience.
Customization
The user is in control and is able to modify content and
the look and feel of content offered on a site.
Personalization
It
is more technology and behavior driven. The site [computer server]
controls what the user sees, based on information about the
user's attributes and behaviors stored on the server.
What
Can Be Customized
Layout:
according to Yahoo experience (see article about Yahoo), people
usually prefers the default page. It has to be consider how
much value is add to user this kind of customization; power
users might use layout customization, but not much the "intermediate
user".
Content:
most effort has been developed to match user preferences. The
goal is "to ensure right people receive the right information
at the right time" [Smith/Lotter]
Personalization
Approaches
1.
Direct manipulation: It is based in user selection
and not automatized.
Advantages
for user: comprehensible, predictable, controllable actions
Disadvantages
for user: does not provide exploration/new options in
case user changes taste or preferences. So it has to be simple,
easy to update, add/remove.
2.
Learn-based customization: It is based on learning
algorithms, also called "self-customizing software". What it
does:
Disadvantages:
unpredictability, such as users might click in an article
or event for curiosity, or for find something that would match
a friend taste, not necessarily his/her taste, and be annoyed
by the options that the system offer.
There are
different strategies and methods:
Content-filtering
method: seeks to recommend similar items for a given user
that similar users also liked. For example, News
Dude use of user feedback about prediction to refine ["interesting
feedback option"]; it considers the preferences of a single
user.
Disadvantages:
problematic and time consuming; limits of the user profile
for future recommendation
Collaborative
filtering method: profiles are based on user assigning
ratings of items/ finding users who assigned similar rates;
some began by asking users to rate something, and cross this
info with other people ratings.
Advantages:
increases with the user bases [this can be a drawback when
the site is launch, so designers might want to wait grown
of user bases to implement it; improve diversity
Disadvantages:
difficulty to deal with "unusual users" who does not fit
in any profile.
Mix
of content and collaborative filtering: The goal here
is to take the advantages of the two systems and improve precision
Research
Questions Underlining Those Methods
- Is it
possible to predict users actions?
- User's
pattern of actions may be different and varies with time
|
