Intelligent Recommender Systems

1
Intelligent Recommender Systems
2
Objectives

• Know what an intelligent recommender systems is
• Understand why it is important in Ecommerce
• Understand the principles of operation of
  intelligent recommender systems

3
Intelligent Recommender Systems

• Examples
• Amazon
• www.MovieFinder.Com
• www.cdnow.com
• Etc.etc.

4
Intelligent Recommender Systems

• What do they do
• Observe behaviour of one set of users
• Assume new users who behave in a similar way to
  an existing group of users will share other
  common behaviours

5
Behaviour

• Explicit Navigation
• More like this
• Implicit Navigation
• Where did they click for more information
• Which categories did they visit
• What did they purchase

6
Operation

• Identify key attributes worth recording
• Navigation
• Searching
• Selection
• Selected Recommendations
• Record them in a vector
• Known as a user or user group profile
• Recommend Products
• On the basis of interactions in user profile

7
Similarity Again

• Vectors recording facts about behaviour
• Effectively queries retrieving outcomes
• Need to think about errors
• What if someone didnt like what was being
  recommended ?
• However fundamentally its another clustering
  problem

8
Book Shop Example

• Things to record
• Books they bought
• Categories they Looked at
• Search Terms they entered
• ? More

9
Problem

• Whats a group
• People who behave similarly to each other
• People who buy similar products ?

10
What are we clustering ?

• People ?

• Products ?

11
Issues

• Privacy
• Turning People off while youre getting to know
  them
• The effect of change
• Balancing personal and global information
• This weeks fashion
• What I like

12
Exercise

• Filumbuff de Noire is setting up a website
  delivering streamed movies on demand.
• He wishes to use an intelligent recommender
  system to recommend new films to returning users.
• Your task is to identify the information from the
  interactions which should be recorded in order to
  make the recommender system as effective as
  possible.

13
Conclusions

• Intelligent Recommender Systems are an important
  aspect of effective Ecommerce
• They operate by clustering people or products
• The big issue is what to record

14
More Information

• John Riedls Web Site
• Middleton, Shadbolt and Roure Ontological User
  Profiling in Recommender Systems ACM Ransactions
  on Information Systems 22(1) 2004 pp 54-88
• Carenini, Smith, and Poole Towards more
  conversational and collaborative recommender
  systems IUI 2003 pp 12-18