Transcript CANTIERE MULTIMEDIALE STET-TELECOM Progetto SERVIZI

Adaptive Web Stores
L. Ardissono, C. Barbero, A. Goy and
G. Petrone
Dipartimento di Informatica
Universita’ di Torino, Torino, Italy
[liliana,cris,goy,giovanna]@di.unito.it
http://www.di.unito.it/~seta
The problem
electronic catalogs are difficult to browse
they often contain very different types of
information, or are not detailed enough
eterogeneous people visit them
people have different interests,
backgrounds, interaction needs
there is no single solution to satisfy all
needs (see also Benyon:93, Smith-etal:97)
May 4, 1999
Agent Architecture for Personalized
Web stores
2
An improvement...
Information Filtering & Electronic
Commerce systems focus on selecting
items suitable to the user’s preferences
(exploiting techniques like collaborative
filtering, case-based reasoning, ...)
An interesting expansion is the focus on
the interactional aspects on the Web
May 4, 1999
Agent Architecture for Personalized
Web stores
3
Our goals
• customization of product descriptions
– presentation of different sets of features
– use of different linguistic descriptions to
present features
– selection of the amount of information to
present (to constrain the information load)
• suggestion of different items of a product
May 4, 1999
Agent Architecture for Personalized
Web stores
4
Personalization strategies in
SETA
To generate the pages our system
– identifies the user preferences and interests
– tailors the contents of the catalog pages to the
user characteristics
– suggests the items best matching the
preferences in the user profile
May 4, 1999
Agent Architecture for Personalized
Web stores
5
Relevant areas
• dynamic hypermedia (to generate Web
pages ‘on the fly’)
• user modeling (to handle user profiles)
• knowledge-based systems (to handle the
information about products and customers)
• distributed agent architectures (to exploit
specialized agents within a complex system)
May 4, 1999
Agent Architecture for Personalized
Web stores
6
Representation of user profiles
• Classification data (age, job, …)
• Personality traits (domain expertise, technical
interest, aesthetic interest, receptivity)
e.g.: Domain Expertise:
<low, 0.9>,<medium,0.1>,<high,0>
• Preferences
e.g.: Ease of use: importance: 1;
<low, 0>,<medium,0.3>,<high,0.7>
May 4, 1999
Agent Architecture for Personalized
Web stores
7
A stereotype (Novice user)
• Classification data:
age: importance: 0.7; <0-24, 0.3>,<25-44,0.2>, ...
job: importance: 0.8; <student, 0.8>,<25-44,0.2>, ...
• Personality traits
domain expertise: <low, 0.9>,<medium,0.1>,<high,0>
technical interest : <low, 0.8>,<medium,0.2>,<high,0>
receptivity: <low, 0.2>,<medium,0.7>,<high,0.1>
• Preferences
ease of use: importance: 0.9 <low, 0>,<med,0.2>,<h,0.8>
quality: importance: 1; <low, 0>,<med,0.6>,<high,0.3>
May 4, 1999
Agent Architecture for Personalized
Web stores
8
Representation of items
VivaVoce T200
• Features
agenda:20 numbers
price: Lit. 90.000
• Properties
ease of use: high
quality: high
• Link to database table
NB: the Features are typed slots (there are technical,
aestetic features, etc.)
May 4, 1999
Agent Architecture for Personalized
Web stores
9
Page tailored to an expert user
May 4, 1999
Agent Architecture for Personalized
Web stores
10
Page tailored to a non-expert user
May 4, 1999
Agent Architecture for Personalized
Web stores
11
Key roles in the architecture I
• Communication with the Web (SessionMgr)
• Management of the interaction flow
(DialogMgr)
• Generation of the catalog pages by applying
personalization strategies (Personalization
agent)
• Initialization and update of user profiles by
applying user modeling acquisition rules
(UMC)
May 4, 1999
Agent Architecture for Personalized
Web stores
12
Key roles in the architecture II
• Selection and rating of the items to suggest
to the user (Product Extractor)
• Management of the Users DB (to maintain
user profiles in a permanent way)
• Management of the Products DB
(containing the information about items)
• Maintenance of the user’s shopping cart
May 4, 1999
Agent Architecture for Personalized
Web stores
13
Matching items to users
• the items to be suggested are scored on the
basis of the preferences in the user profile
• the property values of each item are
matched against the user’s preferences, to
identify the best matching items
• in the scoring process, the importance of the
user’s preferences is exploited to rule out
irrelevant mismatching properties
May 4, 1999
Agent Architecture for Personalized
Web stores
14
The System Architecture
Usrs
DB Mgr
Stereotype
KB
UM-i
Prod
Taxonomy
UMC
Product
Extractor
Dialog
Context
Extr
Context-i
Cart
Products
DBMgr
May 4, 1999
Users DB
Personal
Agent
Dialog Mgr
Shopping
Mgr
W
e
b
Session
Mgr
S
e
r
v
e
r
ProductsDB
Agent Architecture for Personalized
Web stores
15
Three-tier architecture
II level
Solaris JDK 1.1.3
Java Web Server 1.1
Session
Mgr
Agents
I level
W
e
b
S
e
r
v
e
r
Browser_i
Browser _k
Netscape,
Ms Explorer
Users DB
Products DB
NT JDK 1.1.4
ODBC driver
III level
May 4, 1999
Agent Architecture for Personalized
Web stores
16
Conclusions
• SETA: virtual store shell for the
construction of Web stores capable of
tailoring the interaction to the users’ needs
• Agent-based system, where agents have
been associated to each basic role in the
management of the interactions with
customers
• Special attention has been posed on user
modeling and personalization strategies
May 4, 1999
Agent Architecture for Personalized
Web stores
17