TANGENT A Novel, Surpriseme, Recommendation Algorithm

1
TANGENTA Novel, Surprise-me, Recommendation
Algorithm

• Kensuke Onuma Sony Corporation
• Hanghang Tong Carnegie Mellon Univ.
• Christos Faloutsos Carnegie Mellon Univ.

2
Motivation
Movies
Jessica
Liz
John
Jessica
Liz
John
Kevin
Kevin
Tim
Bob
Tim
Bob
Mike
Mike
Mary
Rachel
Mary
Rachel
Tom
Mark
Tom
Mark
Broadening users horizon
More chance to increase sales of items
Go off on a TANGENT !
3
What we want are
target user ( query node)
comedy fans
horror fans
A
user
movie
Conventional recommendation algorithms answer
TANGENTs answer
4
Outline

• Motivation
• Problem definition
• Algorithm
• Experiments
• Conclusion

5
Graphs for recommendationbipartite graph
Mark
Rachel
Tom
Mary
John
Mike
A
B
C
D
E
F
G
H
weighted based on rating
users and movies
6
Problem definition of TANGENT
Given - An edge-weighted undirected graph
with adjacency matrix - The set of query
nodes
user
movie
Find - A node that satisfy following
conditions. (1) Close enough to (2)
Possessing high potential to reach other nodes
7
Outline

• Motivation
• Problem definition
• Algorithm
• Experiments
• Conclusion

8
Outline of TANGENT algorithm

• Calculate relevance score of each node to
• Calculate bridging score of each node
• Compute the TANGENT scoreby merging two criteria
  above

user
movie
9
Step 1 Relevance score
Random walk with restart Pan KDD 04
2
4
5
7
9
query node
1
3
6
8
Various Scalable Solution Tong 06 - OnTheFly
- B_Lin - NB_Lin - BB_Lin (for bipartitle graph)
10
Step 2 Bridging score (Intuition)
a node in a group
a node between groups
2
3
2
5
4
1
6
1
5
3
7
4
7
6
0
0
small
large
11
Step 2 Bridging score (Detail)
2
3
neighbors
1
4
12
Step 3 TANGENT score
A. Simple multiplication. (not linear
combination, not skyline query, )
relevance score among neighbors
query
user
relevance score to query nodes
movie
13
Example
2
4
5
7
9
query node
1
3
6
8
Group 1
Group 2
14
Outline

• Motivation
• Problem definition
• Algorithm
• Experiments
• Synthetic data
• Real data
• MovieLens (user-movie)
• DBLP (author-paper)
• Conclusion

on our paper
15
Synthetic databipartite graph
1
5
2
3
4
6
7
8
9
13
14
15
16
17
18
19
20
21
22
23
10
11
12
24
25
26
16
Real data MovieLens

• User Preference (rating 5)
• - A Nightmare on Elm Street (1984) (Horror)
• The Shining (1980) (Horror)
• Jaws (1975) (Action, Horror)

943 users 1682 movies 55375 ratings
Ranked list by TANGENT score
Ranked list by relevance score
17
User Preference (rating 5) - Robin Hood Men in
Tights (1993) (Comedy) - Young Frankenstein
(1974) (Comedy, Horror) - Naked Gun 33 1/3 The
Final Insult (1994) (Comedy) - Fatal Instinct
(1993) (Comedy)
TANGENT score
relevance score
18
Outline

• Motivation
• Problem definition
• Algorithm
• Experiments
• Conclusion

19
Conclusion

• Definition of a novel recommendation problem
• how to make a recommendation that broadens the
  horizons of the user?
• Approach close to the user preferences
  have high connectivity to other groups
• Design of algorithm
• Relevance score X Bridging score
• Effective Efficient
• Experiments
• synthetic dataset
• real dataset

20
Thank you
Kensuke Onuma Kensuke.Oonuma_at_jp.sony.com
Code available http//www.cs.cmu.edu/kensuke/
Hanghang Tong htong_at_cs.cmu.edu
Poster tonight ! 1930 2200 at Hôtel de Ville
Christos Faloutsos christos_at_cs.cmu.edu