Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 37

Social Recommender

System

By: Ibrahim Sana

15.08.08
Agenda
 Introduction
 Background on Collaborative Filtering
 Collaborative Filtering Limitation
 Using trust in RS
 Related works
 Research methodology
 Evaluation and Results
 Conclusion
Introduction
 Recommender system (RS) help users find items
(e.g., news items, movies) that meet their specific
needs.
 Motivation
 Information overload
 Researches in RS focused on developing methods
and approaches dealing with the Information overload
problem.
 Main Approaches
 Content-Based (Salton, 1989)
 Collaborative filtering/Social Filtering (Goldberg, 1992 )
 hybrid
Collaborative Filtering (CF)
 In the real world we seek advices from our trusted
people
 CF automate the process of “word-of-mouth”
 General use:
 Weight all users with respect to similarity with the active
user.
 Select a subset of the users (neighbors) to use as
predictors (recommenders).
 Rating prediction:
n

w a ,u ( ru ,i  ru )
pa ,i  ra  u 1
n

w
u 1
a ,u
User-User Collaborative Filtering

Active user

3
?

Rating
prediction
CF Limitation
 New item problem
 Cold start problem
 Sparsity (95%-99%)
 Controversial user
 Easy to attacks
 Scalability
 Cannot recommend items to someone with
unique tastes.
 Tends to recommend popular items
Solution: using trust relationships
 Implicit: Deriving trust score directly from the rating
data
 Generally based on user prediction accuracy in the past
 Explicit: users explicitly “rate” other users
 FilmTrust (Hendler et al,2006)
 Molskiing (Massa et al,2005)
 Limitation:
 Users have on average very few links (trusted sources)
 More User’s effort
 Solution
 Trust propagation: find unknown user’s
trustworthiness based on the users’ “web of trust”
Trust inference
 Global metrics: computes a single global trust
value for every single user (reputation) 1 b 5

 Examples: a 3 d
 PageRank (Page et al, 1998),eBuy 2 c 3
 Pros:
 Based on the whole community opinion
 Simple to compute
 Cons:
 Trust is subjective (controversial users)
Local trust metrics
 Local metrics: predicts (different) trust scores that are
personalized from the point of view of every single
user
 Example:
 MoleTrust (Massa et al,2006)
 TidalTrust (Golbeck et al,2005)
 Pros:
1 b 5
 More accurate
? d
 Attack resistance a
 Cons: 2 c 3
 Ignoring the “wisdom of the crowd”
 More complicated
Related works(1):Massa et al(2006)
 Crawling Epinion.com
 users can review items and also assign them numeric
ratings in the range 1 to 5.
 Users can also express their “Web of Trust” and their
Black list
 Dataset:
 ~50K users,~140K items,~665K reviews
 487K binary trust statement
 Sparsity=99.99135%
 Above 50% are cold start users (less than 5 review)
Recommendation method
 Using MoleTrust metric

Input
Estimated trust output
userXuser Predicted Ratings
Rating
MXN
predictor
Rating
MXN

w a ,u ( ru ,i  ru )
pa ,i  ra  u 1
n

w
u 1
a ,u
Evaluation and results
Related works(2):Golbeck et al(2006)
 FilmTrust: Online Recommender System
 Users can rate films, write reviews, and express trust
statements in other users based on how much they trust
their friends about movies ratings
 Rating scale from half start to four start
 Trust scale from 1 to 10
 Dataset:
 500 users, 100 popular movies, 11,250 rating
 350 users with social connection
 Sparsity=77%
Recommendation method
 Weight ratings by trust value
 Search recursively for trusted sources
 Using TidalTrust metric for trust inference
 Simple Prediction method
 Example:
t r
si im

Alice trust Bob 9 rsm  iS

Alice trust Chuck 3


Bob rates the movie “Jaws” with 4 stars t si
Chuck rates the movie “Jaws” with 2 stars iS

Alice’s predicted rating for “Jaws” is: (9*4+3*2)/9+3=3.5


Evaluation and results
 Benchmarks: Pure CF and simple average
 80% training and 20% testing
 Using MAE metric
 First analysis, using trust didn’t appear to be effective
 Above 50% of the rating were within the range of
the mean +/- half star
 Trust-based significantly useful only to user who
disagree with the average
Result
Limitations
 Do not distinguish between various types of social
relationships

 Researches in marketing and in applied psychology


identified different types of social measures impact
recipient’s advice taking

 Different types of social relations impact recipient’s


advice taking in different ways
Dominants Social Measures
 Cognitive similarity (Gilly et al. 1998)
 Tie-Strength (Levin & Cross 2004)
 Relationship duration
 Interaction frequency
 Closeness
 Trust (Smith et al. 2005)
 Competence
 Benevolence
 Integrity
 Social Capital/Reputation (Gilly et al. 1998)
Motivation

 Web 2.0 provide opportunity for peoples to


interact with each other
 Social networks (trust, friendships)
 Electronic communications (Tie-Strength)
 Reputation mechanisms (Social Capital)
Research questions

 Can additional relationship information be


utilized to enhance recommender system
performance?

 What types of social relation is most useful?


Objectives

 Identify the difference between similarity based


CF and social based CF

 Explore the contribution of various social


relations

 Suggest solution for the cold start problem

 Suggest solution for the scalability problem


Hypothesis
 H1:Null Hypothesis: social relationships don’t provide any
contribution to the performance of recommender systems
Alternative Hypothesis: social relationships do contribute to
the performance of recommender systems
 H2:Null Hypothesis: different social relationships provide
different contribution to the performance of recommender
systems.
Alternative Hypothesis: different social relationships provide
similar contribute to the performance of recommender
systems
 H3:Null Hypothesis: different social relationships provide
different contribution to the performance of recommender
systems.
Alternative Hypothesis: different social relationships provide
similar contribute to the performance of recommender
systems
Social dimensions and
measurement
Social dimension Measurement
Trust I trust this person
Friendship I would consider this person a friend
Interaction Frequency How often did you communicate with this person
Relationship Duration How long have you known this person
Social capital This person is reputable
Research Method
 Domain: movie recommendation
 Subject : 97 4th years student from the IS department
(with social relationships)
 Tasks:
 Provide rating for 160 (popular) items (5 point scale)
 Select three subject and indicate your social
relationships
 Some of the relationships we examined
 Trust
 Friendship
 Interaction duration
 Interaction frequency
 Reputation
Research method
Research Method

Independent Variables
(recommendation methods)
Baseline:User-based CF
(Pearson Correlation)
Hybrids method (Similarity
and Social relations) Dependent Variables
(Combination schemes) (Performance)
Social Restriction method
MAE
(Pearson Correlation)

Precision and Recall

Control Variable Coverage


Subjects
Students with social
relationships
Tasks
1-Movies rating
2-Social network building
Experiment Environment
User Authentication

Task1: Movies rating

Task2: User's social relationships


Research framework
Past Ratings

Recipient Sources

System’s Receiver-
Source Similarity
Calculation Systems System’s Prediction
Prediction (Recommendation)
Component

Recipient-Source
similarity
(Recipient’s)
Trust, Sources’
Friendship Qualification
s
Interaction
duration, System’s Source
frequency Qualification
Component
Reputation
Prediction method 1

 Hybrid method
 Social relations combined with similarity (Pearson
Correlation)
 Tuning the source’s weight according to his group
 Group P: sources similar to the active user
 Group S: sources belong to the social network of
the active user
 Pa ,u .......................if (u  P) P
 S
S a ,u .......................if (u  S )
Wa , u  
 f ( Pa ,u , S a ,u )..........if (u  P  S )
0............................Otherwise

Prediction method 2
 Social restriction
 Social relations used for restriction
 Consider only sources belong to both groups
S and P
 Using the source’s similarity

S P
Wa , u  Pa, u.....if (u  P  S )
Simulation System Architecture
Configuration Utility
Front-end

Similarity-Based CF Hybrid CF
Social restriction CF
(Pearson Correlation) (Pearson Correlation
and Social ties)

Offline-Online boundary

User-User Similarity
generation

Fold generation Social Network Propagation


Randomly generate 10 folds
(20% testing, 80% training) (distance 1 to 6)

User-Item Users’ Users’ Users’ Social


Rating Folds Similarity Network
Results (Hybrid method)

Social weighting coverage


Social
Social weighting
weighting
Social Weight Impact
86
0.84
0.7445 0.82
84
0.7440.82 0.8

0.743582
0.8 0.78
0.743 80

Coverage
0.78
0.76

MAE
coverage Precision
MAE 0.742578

MAE
MAE Recall 0.76
0.742 0.74
76
0.74
0.7415
74 0.72
0.7410.72
72 0.7
0.7405
0.7
d=6 d=5 d=4 d=3 d=2 d=1 pure CF 70 0.68
0.74
100d=690 d=5 70 d=5
d=6
80 d=4 d=340method
50d=4
60 Prediction d=2 20 d=2
d=3
30 d=1
10 0pure
d=1 CF pure CF
Prediction
Social method method
Prediction
tie weight
Hybrid method: Cold start users
Impact of shared-interest sources

1.4

1.2

0.8
MAE-CF

MAE
MAE-WAA1
0.6

0.4

0.2

0
25 23 21 19 17 15 13 11 9 7 5 3 1
Number of sources
Impact of different social measures

Social measures AMAE Precision Recall Improvements

Cognitive similarity 0.822165 0.798522 0.731773  

Tie-Strength 0.746838 0.795328 0.749967 9.162064

relationship duration 0.742796 0.796085 0.746972 9.65368

interaction frequency 0.748337 0.795426 0.750318 8.979698

Closeness 0.74938 0.794472 0.752611 8.852814

Trust 0.738412 0.796603 0.743428 10.18684322

competence 0.738428 0.797798 0.741061 10.1849155

benevolence 0.736044 0.795776 0.744768 10.474944

Integrity 0.741934 0.795456 0.746152 9.7585475

Social capital 0.744079 0.797383 0.74192 9.497685


Result (Social restriction)

Socialrestriction
Social restriction
Social restrictin coverage
0.84
0.82
80
0.8
0.82
70
0.78
60
0.8
0.76
50
Precision 0.78

Coverage
0.74

MAE
MAE
Recall
coverage 40
0.76
0.72
30
0.7
0.74
20
0.68
0.72
10
0.66
d=6 d=5 d=4 d=3 d=2 d=1 0
pure CF 0.7
d=6 d=6 d=5 d=5 d=4 d=3
d=4 Prediction
d=3 method d=2
d=2 d=1
d=1 pureCF
pure CF
Prediction method
Prediction method
Social restriction: cold start users

Shared interest sources impact

2.5
2

MAE-CF 1.5

MAE
MAE-WAA1 1

0.5

0
25 23 21 19 17 15 13 11 9 7 5 3 1
Number of sources
Conclusion
 Social relationships is effective in alleviating
CF weaknesses:
 Cold start problem (Social weighting and
social restriction)
 Scalability problem (Social restriction)
 Spammers attacks (Social weighting and
social restriction)
References
 Shardanand, U., Maes, P.: Social Information Filtering: Algorithms for Automating ’Word of
Mouth’. In: Proceedings of Human Factors in Computing Systems, pp.10–217 (1995)

 Herlocker, J., Konstan, J.A., Terveen, L., Riedl, J.: Evaluating Collaborative Filtering
Recommender Systems. ACM Transactions on Information Systems 22, 5–53(2004)

 Massa, P., Avesani, P.: Trust-Aware Collaborative Filtering for Recommender Systems.In:
Proceedings of the International Conference on Cooperative Information Systems (CoopIS),
Agia Napa, Cyprus, pp. 492–508 (2004)1060 C.-S. Hwang and Y.-P. Chen

 Avesani, P., Massa, P., Tiella, R.: Moleskiing: A Trust-Aware Decentralized Recommender
System. In: Proceedings of the First Workshop on Friend of a FriendSocial Networking and
the Semantic Web, Galway, Ireland (2004)

 Golbeck, J: Generating Predictive Movie Recommendations from Trust in Social Networks.


Proceedings of the Fourth International Conference on Trust Management. Pisa, Italy, May
2006.

 R. Guha, R. Kumar, P.:Raghavan, and A. Tomkins. Propagation of trust and distrust. In


Proc. of the Thirteenth International World Wide Web Conference, MAY 2004.

You might also like