Recommendation Explorer

1
Recommendation Explorer

• Using Dimensionality Reduction to Improve
  Similarity Judgements

2
Recommendation Explorer

• Using Dimensionality Reduction to Improve
  Similarity Judgements

Overview of RecEx Current Research item similari
ty module Description of current module Evaluati
on/discussion of module performance
Conclusion
3
Recommendation

• Personal
• Contextual

4
Recommendation

• Personal
• Contextual

5
Item Similarity Module Data

• 12,726 popular films

Production info (director, year, etc.)
Plot summary
Genre
6
Item Similarity Module Data

• 12,726 popular films

Editors Picks -- 0 or more close recommenda
tions
-- 0 or more creative recommendations
7
(No Transcript)
8
Implied Relationship
Terminator
Star Wars
Alien
9
Singular Value Decomposition
S
T
D
A

• m x n m x r
  r x r r x n

10
Singular Value Decomposition
Sk
Dk
Ak

Tk

• m x n m x k
  k x k k x n

11
RecEx Similarity Model

• Sim(filmu, filmv)

That Tk Sk
In the space defined by That where
12
Experimental Evaluation

• N 133 volunteer reviewers
• 10 seed films

13
Experimental Evaluation

• Precision at 5 levels of recall
• (.1, .25, .4, .5, .75)
• Used only for the Full Key

14
Experimental Evaluation

• Weighted avg. search length
• Used with all 3 keys
• Not an entirely fair metric for raw method

15
Experimental Evaluation
Average Precision of recommendations based on raw
link structure and links analyzed by SVD

16
Experimental Evaluation
Hypothesis tests concerning the equality of mean
weighted ASL for raw and SVD-derived
recommendations

17
Experimental Evaluation
Hypothesis tests concerning the equality of mean
weighted ASL for raw and SVD-derived
recommendations

18
Experimental Evaluation
Hypothesis tests concerning the equality of mean
weighted ASL for raw and SVD-derived
recommendations