A preference scoring technique for personalized advertisements on Internet storefronts

1
A preference scoring technique for personalized
advertisements on Internet storefronts

• Jong Woo Kim, Kyung Mi Lee, Michael J. Shaw,
• Hsin-Lu Chang,
• Matthew Nelson, Robert F. Easley
• Mathematical and Computer Modelling 44 (2006)
• Page 315

2
Introduction

• The methods proposed here require only a small
  amount of data for a single customer, and involve
  very simple updating rules and selection
  algorithms
• Providing a great advantage in performance and
  minimizing privacy concerns, which are both key
  issues in this market

3
Introduction

• Rule-based approach it may not be effective to
  show the same advertisements to all the customers
  satisfying some conditions
• Collaborative filtering a serious limitation is
  that a new customer has to provide preferences
  for a large number of items in order to view
  personalized advertisements

4
Preference Score

• Without rely on customer clustering, or use
  similar customers opinions
• personalized recommendation is achieved using
  only the information of each customer without
  reference to a customers demographic data or
  other personal information
• Not require learning data sets or learning phases
• Reflect changes in customers preference

5
Structure to provide real-time personalized
advertisements
6
Models and algorithms
7
Models and algorithms
8
Preference Tables

• Stores customer preference scores
• Customer ID (CID), Product Group ID (PGID),
  Preference Score (PS)

• Where PS(i, j ) is the preference score of
  customer i for the leaf-level product category j
• Update of Preference Scores

9
(No Transcript)
10
Preference Tree

• Preference table cannot reflect affinity levels
  among product categories

11
Preference Tree

• The above update procedure does not require the
  update of all preference scores in the tree,
  rather it requires only the update of the
  preference scores of the related leaf node and
  its ascendant nodes