Yoda an accurate and scalable web based recommendation systems

Yoda: An Accurate and
Scalable Web-based
Recommendation Systems
Cyrus Shahabi, Farnoush Banaei-Kashani,
Yi-Shin Chen, and Dennis McLeod
Integrated Media Systems Center and
Computer Science Department,
University of Southern California
E-mail:{shahabi, banaeika, yishinc, mcleod}@usc.edu

Outline
 Motivation
 Related Work
 Content-based Filtering
 Collaborative Filtering
 Offline Process: Clustering, Voting, Aggregation
 Online Process: Classification & Aggregation
 Performance Evaluation
 Conclusion & Future Work

Motivation
 The amount of data is enormous on the Web
 Users suffer from information overload
 Recommendation systems can personalize and customize
the Web environment in real-time
 Similar to Amazon.com “real-time” recommendations (people who bought
this book also purchased …)
 Different approach (vs. association-rule mining)
 Challenges:
 Scalability : As the # of items and users grow, the system stay efficient
 Sparsity: Not enough information available on the user

Related Work:
Content-Based Filtering
 From the Information Retrieval community [Maes1994]
[Shardanand and Maes 1995] [Balabanovi and Shoham 1997]
 Based on a comparison between the feature vectors of
items (e.g., artist, style) in the database and the user’s
interest list
 Major weakness [Balabanovi and Shoham 1997]
 Content limitation: only can be applied to few kinds of content, can
only capture certain aspects of the content
 Over-specialization: users can only obtain information based on the
content of their profiles

Related Work:
Collaborative Filtering(CF)
 Employ a user’s item evaluations (not the actual content)
to find other similar users: nearest-neighbor algorithm
[Resnick et al. 1994]
Three major weaknesses
 Scalability: time complexity O(U*I) (I:#items, U: #users)
 Clustering [Breese et al. 2000]
 Bayesian network [Kitts et al. 2000]
 Sparsity: profile matrix (i.e., # of user evaluated items) is sparse
 SVD [Sarwar et al. 2000]
 Synonymy: latent association between items is not considered
 Content analysis [Balabanovi and Shoham 1997]
 Categorization [Kohrs and Merialdo 2000]

Fuzzy
Aggregation
Fuzzy
Aggregation
Clusters
Offline Process
PPED
Similarity
Measure
and
Clustering
PPED
Similarity
Measure
and
Clustering
User Navigation Behaviors
User 1
User 2
User 3
User 4
User 5
User U-6
User U-5
User U-4
User U-3
User U-2
User U-1
User U
User 6
VotingVoting
Favorite
PVs
(Rock=
High
Classical=
Low
Pop=
Low
Rap=
High)
Item Database
Cluster
Wish-list
0.87
0.83
0.72
0.47
0.61

Voting Mechanism
Favorite
PVs
(Rock=
High
Classical=
Low
Pop=
Low
Rap=
High
Blues=
Low)
Rock Classical Pop Rap Blues
High Low Mid High Low
Property Values
VotingVoting
RockClassicalBlues
HMLHMLHML
51221071561212537
Cp,f(k)
Mpf=Max{Cp,f(k)}
f in F
( ) ( ){ }pffpp MkCFfffmaxkF =∈= ,,

Ranking Items
Item Database
Cluster
Wish-List
0.87
0.83
0.82
0.79
0.72
0.70
0.68
0.65
0.63
0.61
0.54
0.47
0.42
Fuzzy
Aggregation
Fuzzy
Aggregation
( ) ( ){ }kFpfmaxiv pik ×= ~
fmax{ …}
Favorite
PVs
(Rock=
High
Classical=
Low
Pop=
Low
Rap=
High
Blues=
Low)
Fp(k)
(High*High) , (Mid*Low), (Low*Low)
Vk(i)
High Low Mid Mid Low
Property Values
ip~
Locality Sensitive
Hashing algorithm

Favorite
PVs
(Rock=
High
Classical=
Low
Pop=
Low
Rap=
High
Blues=
Low)
High Low Mid Mid Low
Property Values
Fuzzy
Aggregation
Fuzzy
Aggregation
fmax{ }
( ) ( ){ }ϑ∈∀= fiEfmaxiv fkk ,
Optimized Equation
 Why optimized: time complexity O(#P*I) (#P: # of properties, I: # of items)
 Intuition: the vk(i) value comes from the maximum value among ( ){ }ip pkF ~×
Mhigh(k) ( ){ }kGppfmaxkM fif ∈= ~)(
( ) )(, kMfiE ffk ×=
f
(High*High), (Low*Mid)

Optimized Equation
 Optimized Equation

 Time complexity: O(f*I) I=#items f=#fuzzy terms
 Satisfy a triangular norm form
 Time complexity can be further reduced to O(N) (N: constant number) by
Fagin’s A0 algorithm [Fagin 1996]
( ){ }kGppfmaxkM fif ∈= ~)(
( )
( ) ( ){ }ϑ∈∀=
×=
fiEfmaxiv
kMfiE
fkk
ffk
,
, )(

PPED
Similarity
Measure
PPED
Similarity
Measure
Fuzzy
Aggregation
Clusters
Online Process
Current User’s
Navigation Behavior
A List of Similarity Values
0.65 0.790.32
User
Wish-List
0.87
0.83
0.82
0.79
0.72
0.70
0.68
0.65
0.63
0.61
0.54
0.47
0.42
Cluster Wish-lists
0.87
0.83
0.72
0.47
0.61
0.87
0.83
0.72
0.47
0.61
0.87
0.83
0.72
0.47
0.61

Optimized Method
 Original Time complexity: O(K*I) K=#clusters I=#items
 Time complexity of optimized method:
 O(f*I) f=#fuzzy terms
 Time complexity can be further reduced to O(N) (N:
constant number) by Fagin’s A0 algorithm [Fagin 1996]

Experimental Methodology
Clustering
Generate
Item Database User Set
Clusters
Similarity Matrix
cluster
user
Cluster Favorite PVs
Ranking
of Items
in Clusters
Generate
Assign Property Values
to Items:
• Item-PV =
f(Cluster-PV, noise)
• noise ~ item-rank

Clusters
Similarity Matrix
cluster
user
Cluster Favorite PVs
Ranking
of Items
in Clusters
H L M N F F L
L M N F F L
M N F F L M
N F F L
M N F F L
L M N F F L
Assign evaluation values to items
•Item-Rating = f(Cluster-Ranking, weight)
• weight ~ user-cluster similarities

H L M N F F L
L M N F F L
M N F F L M
N F F L
M N F F L
L M N F F L
Training
Testing
Current Session Recommendation

Accuracy Comparison
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
1000 5000
Number of Items
HarmonicMean
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
Improvement
Nearest Neighbor Method Yoda Improvementrecallprecision
11
2
MeanHarmonic
+
=

Processing Time Comparison
0
500
1000
1500
2000
2500
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Number of Users
CPUTime(milliseconds/user)
Yoda BNN: Basic Nearest Neighbor Method
Processing Time= CPU +IO
In BNN process: #Items = 5000; #Users = 1000
In Yoda process: #Items in each cluster wish-list = 250 #Clusters = 18

Conclusion
 Yoda scales as the # of users/items grow
 Higher accuracy
Future Work
 Compare other techniques
 Run more experiments with real data
 Incorporate the content-based filtering mechanism into
the user clustering & classification phases
 Incorporate the user profiles

Reference
 [Shardanand and Maes 1995] U. Shardanand and P. Maes, Social Information
Filtering: Algorithm for automating ''Word of Mouth'', proceedings on Human factors in
computing systems, Denver,CO,USA , p. 210-217, May, 1995
 [Maes 1994] Pattie Maes, Agents that reduce work and information overload,
Communications of the ACM, 37(7), p.30-40, 1994
 [Balabanovi and Shoham 1997]Marko Balabanovi and Yoav Shoham, Fab: content-
based, collaborative recommendation, Communications of the ACM, 40(3), p. 66-72,
1997
 [Resnick et al. 1994] P. Resnick and N. Iacovou and M. Suchak and P. Bergstrom and
J. Riedl, GroupLens: An Open Architecture for Collaborative Filtering of Netnews,
Proceedings of ACM conference on Cumputer-Supported Cooperative Work, Chapel
Hill, NC, p.175-186, 1994
 [Sarwar et al. 2000] B. Sarwar and G. Karypis and J. Konstan and J.Riedl, Application
of Dimensionality Reduction in Recommender System -- A Case Study, ACM WebKDD
2000 Web Mining for E-Commerce Workshop, 2000
 [Kohrs and Merialdo 2000] A. Kohrs and B. Merialdo, Using category-based
collaborative filtering in the Active WebMuseum, Proceedings of IEEE International
Conference on Multimedia and Expo, 1, p.351-354, 2000

Reference
 [Kitts et al. 2000] Brendan Kitts and David Freed and Martin Vrieze, Cross-sell: a fast
promotion-tunable customer-item recommendation method based on conditionally
independent probabilities, Proceedings of the sixth ACM SIGKDD international
conference on Knowledge discovery and data mining, Boston, MA USA, p. 437-446,
August, 2000
 [Breese et al. 2000] J. Breese and D. Heckerman and C. Kadie, Empirical Analysis of
Predictive Algorithms for Collaborative Filtering, Proceedings of the Fourteenth
Conference on Uncertainty in Artificial Intelligence, Madison, WI USA, p.43-52, July,
1998
 Shahabi C., A.M. Zarkesh, J. Adibi, and V. Shah: Knowledge, Discovery from Users
Web Page Navigation, Proceedings of the IEEE, RIDE97 Workshop, April, 1997.
 Shahabi C., F. Banaei-Kashani, J. Faruque, and A. Faisal: Feature Matrices: A Model
for Efficient and Anonymous Web Usage Mining , EC-Web 2001, Germany, September
2001
 Fagin R.: Combining Fuzzy Information from Multiple Systems, Proceedings of
Fifteenth ACM Symposyum on Principles of Database Systems, Montreal, pp. 216-226,
1996.
 Shahabi C., and Y. Chen: A Unified Framework to Incorporate Soft Query into Image
Retrieval Systems , International Conference on Enterprise Information Systems,
Setubal, Portugal, July 2001

Yoda an accurate and scalable web based recommendation systems

More Related Content

Viewers also liked (10)

Similar to Yoda an accurate and scalable web based recommendation systems (20)

Recently uploaded (20)

Yoda an accurate and scalable web based recommendation systems