WISECON: The Intelligent Support for E-commerce

1
WISECON The Intelligent Support for E-commerce

• Petr Berka, Tomáš Kocka, Tomáš Kroupa Laboratory
  for Intelligent Systems
• University of Economics, Prague

2
Intelligent support for Internet shopping

• help the user to decide which products to buy,
• find specifications and reviews of the products,
• make recommendations,
• find the best price for the desired product
  (comparison shopping),
• monitor new products on the product list,
• watch for special offers or discounts.

3
Recommender systems

• Recommender systems use product knowledge
  either hand-coded knowledge provided by experts
  or mined knowledge learned from the behavior of
  customers to guide customers through the
  often-overwhelming task of locating products they
  will like.
• Schafer J.B., Konstan J.A., Riedl,J. E-Commerce
  Recommendation Applications. Data Mining and
  Knowledge Discovery 5 (2001) 115-153.

4
E-commerce Recommender application models (1/2)

• Broad recommendation list
• Helping new and infrequent visitors
• (no personal info needed)
• Customer comments and ratings
• Building credibility through community
• (one-to-one marketing)
• Notification services
• Inviting customers back

5
E-commerce Recommender application models (2/2)

• Product associated recommendations
• Cross-selling
• Deep personalization
• Building long-term relationships

6
Example amazon.com customer who bought

• Product associated model (cross-selling)
• Recommendation method item-to-item correlation
• Customer input implicit navigation
• Community input purchase history
• Output suggestion
• Ephemeral personalization
• Passive delivery of unordered list

7
Intelligent Shopping Assistant WISECON - overview
(1/2)

• WISECON - Support of access to on-line catalogue
  of IBM PCs
• Browsing/search
• Clustering products
• Recommending
• Experts knowledge
• Community input

8
Intelligent Shopping Assistant WISECON - overview
(2/2)

• Broad recommendation model
• Recommendation method attribute-based
• Customer input implicit navigation
  keyword/item attributes
• Community input purchase history
• Output suggestion
• Personalization none to ephemeral
• Passive delivery of ordered list

9
WISECON Inference Cycle

• Improve browsing of the on-line catalogue
• Recommend products
• Control the communication with the user

10
Clustering of Products

• To make both browsing and recommending more
  comprehensible

11
Requirements on recommending module

• use expert knowledge
• easy to update to new products
• reflect technological development
• accept vague requests

12
Possible methods

• Bayesian network
• (EUNITE01, ISMIS02)
• Possibilistic network
• (SCI02, IEEE IS02)
• (expert system, CBR, ... )

13
Possibility vs. Probability
14
WISECON Network
15
Interaction with the User

• Considering various types of the user
• Expert
• Middle experienced
• Inexperienced
• (this information is given by the user)
• Asking only such questions that have the main
  impact on discrimination between computers
• selection of questions based on mutual
  information (in probability) or interactivity
  measure (in possibility)