Personalization Using Collaborative Filtering

1
PersonalizationUsing Collaborative Filtering
Fall 2003 Data Mining
2
Personalization

• The building of models that predict a customers
  interest\likelihood to purchase a product.
• Models are used to make customized offerings

3
Recommender Systems

• Systems that build models of customer preferences
  from data, in order to generate personalized
  recommendations.

4
Approaches to Personalization

• Traditional approach to personalization
• Models that use information about consumers
  demographics, purchasing behavior, etc to predict
  their response.
• Example
• Assume a system generating restaurant
  recommendations to customers
• The system predicts the rating a customer would
  assign to a restaurant and uses the ratings to
  make effective recommendations

5
Example Personalization

• Build a regression model to predict ratings for a
  particular customers
• Possible predictors
• Cuisine
• Price range
• Kids
• For a particular restaurant
• Ratinga1(Italian)a2(Japaneese)b(average_cost)
  c(kids_friendly)..e
• However, difficult to understand tacit predictors
  pertaining to consumers tastes and judgment
• E.g., Atmosphere, crowd, etc.
• Critical indicators that are tacit and very
  difficult to tap into

6
Collaborative Filtering

• An attempt to tap into tacit knowledge about
  human taste and judgment
• Predict consumers tastes without having to
  understand and gather information about the
  underlying predictors
• How does it work?
• What products is it likely to work best for?
• What are the potential drawbacks for sellers?

7
Collaborative Filtering The Main Idea

• One seeks recommendations about movies,
  restaurants, books etc. from people with similar
  tastes
• Automate the process of "word-of-mouth" by which
  people recommend products or services to one
  another.

8
Collaborative Filtering The Main Idea

• A customer is characterized by the ratings the
  customer assigned to restaurants he/she dined in
• A customer then receives recommendations from
  customers who assigned similar ratings to
  restaurants.

9
Collaborative filtering The Inner working

• Consumers preferences are registered

• David is seeking recommendations on restaurants
• Using a similarity metric, a subgroup of people
  is selected whose preferences (i.e., restaurant
  ratings) are similar to Davids
• Their (weighted) average ratings for any given
  restaurant is computed, and restaurants with a
  high average score are recommended to David

10
What Algorithms Are Used to Implement
Collaborative Filtering?

• K-Nearest Neighbors

11
Collaborative Filtering ExampleUsing 3-Nearest
Neighbor
(545)/3 4.66
12
Collaborative Filtering

• David receives recommendations from like minded
  customers
• The ratings implicitly embed information about
  tastes, esthetics and human judgment

Swiss 5
Swiss 5
Swiss 1
David
Swiss 4
Swiss 2
13
Who Uses Collaborative Filtering?
14
Example
Product rating
15
Collaborative Filtering Drawback for sellers

• Works well only once a "critical mass" of
  preference has been obtained
• Need a very large number of consumers to express
  their preferences about a relatively large number
  of products.

16
Collaborative Filtering Drawback for sellers

• Consumer input is difficult to get
• Solution identify preferences that are implicit
  in people's actions
• For example, people who order a book implicitly
  express their preference for the book they buy
  over other books
• Works well but results are not as good as with
  the use of rating.

17
Example
Implicit rating
18
Collaborative Rating Weighting The Inputs of
Nearest Neighbors

• Idea Some close neighbors are closer than
  others. Should ratings of all nearest neighbors
  be weighted equally?
• Instead of calculating the average rating among
  closest neighbors
• Calculate weighted average Assign higher
  weights to ratings of closer neighbors

Rachel
John
Hiro 5
Hiro 3
Hillary
David
Hiro 2
19
Collaborative Rating Weighting The Inputs of
Nearest Neighbors

• Weights are disproportional to the distance
  1/0.5 1/1 1/0.2 (2, 1, 5)
• Sum of weights 2158
• Normalized weights 2/8 1/8 5/8
• Hero rating (5/8)5(2/8)2(1/8)34
• 62.5 of the rating is influenced by Rachels,
  25 by Hillarys and 12.5 by Johns

Rachel
John
(distance0.2)
Hiro 5
Hiro 3
(distance1)
Hillary
David
Hiro 2
(distance0.5)
20
Distance with Nominal Variable
Create dummy variables Distance (John,
Rachel)sqrt (1-1)2(1-0)2 (1-0)2
John Married1 (yes) HouseOwner0
(No) CarOwner1 (Yes)
Rachel Married1 (Yes) HouseOwner1
(Yes) CarOwne 0 (No)