Title: Evaluating Similarity Measures: A Large-Scale Study in the orkut Social Network
1Evaluating Similarity Measures A Large-Scale
Study in the orkutSocial Network
- Ellen Spertus
- spertus_at_google.com
2Recommender systems
- What are they?
- Example Amazon
3Controversial recommenders
- What to do when your TiVo thinks youre gay,
Wall Street Journal, Nov. 26, 2002
http//tinyurl.com/2qyepg
4Controversial recommenders
- What to do when your TiVo thinks youre gay,
Wall Street Journal, Nov. 26, 2002
http//tinyurl.com/2qyepg
5Controversial recommenders
- What to do when your TiVo thinks youre gay,
Wall Street Journal, Nov. 26, 2002
http//tinyurl.com/2qyepg
6Controversial recommenders
- Wal-Mart DVD recommendations
http//tinyurl.com/2gp2hm
7Controversial recommenders
- Wal-Mart DVD recommendations
http//tinyurl.com/2gp2hm
8Controversial recommenders
- Wal-Mart DVD recommendations
http//tinyurl.com/2gp2hm
9Googles mission
- To organize the world's information and make it
universally accessible and useful.
10 communities
11Community recommender
- Goal Per-community ranked recommendations
- How to determine?
12Community recommender
- Goal Per-community ranked recommendations
- How to determine?
- Implicit collaborativefiltering
- Look for common membership between pairs of
communities
13Terminology
- Consider each community to be a set of members
- B base community (e.g., Pizza)
- R related community (e.g., Cheese)
- Similarity measure
- Based on overlap BnR
14Example Pizza
15Example Pizza
16Terminology
- Consider each community to be a set of members
- B base community (e.g., Wine)
- R related community (e.g., Linux)
- Similarity measure
- Based on overlap BnR
- Also depends on B and R
- Possibly asymmetric
17Example of asymmetry
18Similarity measures
- L1 normalization
- L2 normalization
- Pointwise mutual information
- Positive correlations
- Positive and negative correlations
- Salton tf-idf
- Log-odds
19L1 normalization
- Vector notation
- Set notation
20L2 normalization
- Vector notation
- Set notation
21Mutual information positive correlation
- Formally,
- Informally, how well membership in the base
community predicts membership in the related
community
22Mutual information positive and negative
correlation
23Salton tf-idf
24LogOdds0
- Formally,
- Informally, how much likelier a member of B is to
belong to R than a non-member of B is.
25LogOdds0
- Formally,
- Informally, how much likelier a member of B is to
belong to R than a non-member of B is. - This yielded the same rankings as L1.
26LogOdds
27Predictions?
- Were there significant differences among the
measures? - Top-ranked recommendations
- User preference
- Which measure was best?
- Was there a partial or total ordering of measures?
28Recommendations for I love wine (2400)
29Experiment
- Precomputed top 12 recommendations for each base
community for each similarity measure - When a user views a community page
- Hash the community and user ID to
- Select an ordered pair of measures to
- Interleave, filtering out duplicates
- Track clicks of new users
30Click interpretation
31Click interpretation
32Overall click rate (July 1-18)
Total recommendation pages generated 4,106,050
33Overall click rate (July 1-18)
34Overall click rate (July 1-18)
35Analysis
- For each pair of similarity measures Ma and Mb
and each click C, either - Ma recommended C more highly than Mb
- Ma and Mb recommended C equally
- Mb recommended C more highly than Ma
36Results
- Clicks leading to joins
- L2 MI1 MI2 IDF L1 LogOdds
- All clicks
- L2 L1 MI1 MI2 IDF LogOdds
37Positional effects
- Original experiment
- Ordered recommendations by rank
- Second experiment
- Generated recommendations using L2
- Pseudo-randomly ordered recommendations, tracking
clicks by placement - Tracked 1.3 M clicks between September
22-October 21
38Results single row (n28108)
Namorado Para o Bulldog
39Results single row (n28,108)
p.12 (not significant)
40Results two rows (n24,459)
41Results two rows (n24,459)
p lt .001
42Results 3 rows (n1,226,659)
43Results 3 rows (n1,226,659)
p lt .001
44Users reactions
- Hundreds of requests per day to add
recommendations - Angry requests from community creators
- General
- Specific
45Amusing recommendations
C
46Amusing recommendations
C
Whats she trying to say? For every time a woman
has confused you
47Amusing recommendations
Chocolate
48Amusing recommendations
Chocolate
PMS
49Allowing community owners to set recommendations
50Allowing community owners to set recommendations
51Manual recommendations
- Eight days after release
- 50,876 community owners
- Added 267,623 recommendations
- Deleted 59,599 recommendations
- Affecting 73,230 base communities and
- 111,936 related communities
- Open question How do they compare with automatic
recommendations?
52Future research 1
- Determining similar users based on common
communities - Is it useful?
- Will the measures make the same total order?
(9 users)
53Other types of information
- Distance in social network
- Demographic
- Country
- Age
- Etc.
54Future research 2
- Per-user community recommendations
- Using social network information
- Using profile information (e.g., country)
55Future research 2
- Per-user community recommendations
- Using social network information
- Using profile information (e.g., country)
56Future research 2
- Per-user community recommendations
- Using social network information
- Using profile information (e.g., country)
57Future research 3
- Do we get the same ordering for other domains?
L2 MI1 MI2 IDF L1 LogOdds
58Acknowledgments
- Mehran Sahami
- Orkut Buyukkokten
- orkut team
59Bonus material
60Self-rated beauty
- beauty contest winners
- very attractive
- attractive
- average
- mirror-cracking material
61Self-rated beauty men
- beauty contest winners 8
- very attractive 18
- attractive 39
- average 24
- mirror-cracking material 11
62Self-rated beauty women
- beauty contest winners 8
- very attractive 16
- attractive 39
- average 27
- mirror-cracking material 9
63Self-rated beauty by country
- Most beautiful
- men
- women
- Least beautiful
- men
- women
64Self-rated beauty by country
- Most beautiful
- men Syrian
- women Barbadian
- Least beautiful
- men Gambian
- women Ascension Islanders
65Ratings by others
- Karma
- trustiness
- sexiness
- coolness
- How do these correlate with age?
66Ratings by others
67Friend counts
68Self-rated best body part