trust and recommender systems

30
ZhaYef 2013.6 1 Trust and Recommender System

Upload: zhayefei

Post on 05-Dec-2014

538 views

Category:

Education


5 download

DESCRIPTION

A brief introduction of the knowledge of Trust and Recommender systems.

TRANSCRIPT

Page 1: Trust and  Recommender Systems

1

ZhaYefei

2013.6.24

Trust and Recommender System

Page 2: Trust and  Recommender Systems

2

Outline

Recommender SystemTrust ModelsTrust in Recommender SystemConclusion

Page 3: Trust and  Recommender Systems

3

Recommender System

Information overload Classified catalogue Search

Ask for friends Two-win

Info Producer Info Consumer

Benefit Long tail 2/8

Why ?

Page 4: Trust and  Recommender Systems

4

Recommender System

Application

Amazon

More than 35% sale are from Recommender System!

Rating

Explaination

Page 5: Trust and  Recommender Systems

5

douban FM

hulu Like ?

60% users benefit!

Recommender System

Page 6: Trust and  Recommender Systems

6

Recommender System

Collaborative Filtering

Content-based Filtering

Algorithm

Item-basedUser-based

1st 2nd 3rd

Page 7: Trust and  Recommender Systems

7

Recommender System

Content-based

Filtering

Movie A

Movie B

Movie C

Like

Like

Like

Movie AType :

Love; Romantic

Movie BType :

Horror;Thriller

Movie CType :

Love; Romantic

similar

User A

User B

User C

Page 8: Trust and  Recommender Systems

8

User-basedFiltering

Recommender System

Item A

Item B

Item C

Item D

Like

Recommend

User A

User B

User C

Page 9: Trust and  Recommender Systems

9

Item-based Filtering

Item A

Item B

Item CLike

Recommend

similar

Recommender System

User A

User B

User C

Page 10: Trust and  Recommender Systems

10

Local Trust

PageRank

Models

Mole Trust Tidal Trust

1st 2nd 3rd

Trust

Global Trust

Page 11: Trust and  Recommender Systems

11

Paolo Massa Italy SAC 2005

(Symposium on Applied computing. ACM, 2005)

A Trust-enhanced Recommender System application: Moleskiing

MoleTrust

Page 12: Trust and  Recommender Systems

12

MoleTrust

G

H

I

A

B

C

D

E

F

0 1 2 3

dist

0 A

1 B C D

2 E F

3 G H I

Page 13: Trust and  Recommender Systems

13

MoleTrust

A

B

C

D

E

F

Setp1 --(BFS) dist=0,1,2 user[dist] user[dist-1]

dist=0, user[0]= Adist=1, user[1]=B,C,Ddist=2, user[2]=E,F

Setp2 trust(A)=1 For each dist =1,2,…

Page 14: Trust and  Recommender Systems

14

( )

( )

( ( )* ( , ))

( )( )

i pre u

i pre u

trust i edge i u

trust utrust i

Setp2 For each u in user[dist]

trust(i=pre(u)) >=0.6

eg.

A

B

C

D

E

F

0.8

0.7

0.5

0.8

0.7

0.7

0.8

dist=1 : Trust(B)=0.8; Trust(C)=0.7; Trust(D)=0.5;dist=2: Trust(E)=(0.8*0.6+0.7*0.7)/(0.8+0.7)=0.65 Trust(F)= (0.7*0.7)/0.7=0.7

MoleTrust

Page 15: Trust and  Recommender Systems

15

Jennifer Ann GolbeckUniversity of Maryland Ph.D thesis 2005

Computing and Applying Trust in Web-base Social Networks

TidalTrust

Page 16: Trust and  Recommender Systems

16

TidalTrust

G

H

I

A

B

C

D

E

F

( )

| ( ) |

jsj adj i

is

t

tadj i

1st

: the trust rating from node i to node jijteg.

2AB AC

AE

t tt

2AE AF

AG

t tt

( )is jst f t

Page 17: Trust and  Recommender Systems

17

TidalTrust

G

H

I

A

B

C

D

E

F

: the trust rating from node i to node j

ijt

2nd

( )

( )

ij jsj adj i

isij

j adj i

t t

tt

3rd

( ) max

( ) max

ij

ij

ij jsj adj i t

isij

j adj i t

t t

tt

Page 18: Trust and  Recommender Systems

18

TidalTrust

Sc

9 8 10

9 9

Sk

8 6

8

89

9 91010

9

Choose The Max as Threshold

2nd

Maximum

9 8 10

8910

9

1st Min=8Min=8 Min=9

9

Setp1 --(BFS)

Page 19: Trust and  Recommender Systems

19

TidalTrust

Sc

9 8 10

9 9

Sk

8 6

8

89

9 91010

9

Choose The Max as Threshold

The shortest path Num=3 Setp2

Max( Strength Paths to Sink )

Max(9,9)=9

Page 20: Trust and  Recommender Systems

20

MoleTrust VS. TidalTrust

G

H

I

A

B

C

D

E

F

MoleTrust: Trust(AG) => Trust(AE)Trust(EG)

AB E

GTidalTrust: Trust(AG) => Trust(AB)Trust(BG)

A

B EG

Page 21: Trust and  Recommender Systems

21

PageRank

A

C

B

D

E

1

1

( )( )( ) (1 ) ( ... )

( ) ( )n

n

PR tPR tPR A d d

C t C t

eg.( ) ( )

( ) (1 0.85) 0.85*( )1 3

PR B PR CPR A

1..

1..

( )* ( )( ) (1 ) ( )

( )

i ii n

ii n

C t RP tRP A d d

C t

?

Page 22: Trust and  Recommender Systems

Trust-aware Recommender Systems

Trust in Recommender Systems

Paolo Massa ItalyRecSys2007

John O’Donovan University College Dublin(Ireland) IUI2005

(International Conference on Intelligent User Interfaces)

Trust in Recommender System

Page 23: Trust and  Recommender Systems

23

Trust

Trust in Recommender System

Collaborative FilteringData sparsityBe easily attacked

Page 24: Trust and  Recommender Systems

24

Trust in Recommender System

( )

( )

( , )( ( ) )

( )| ( , ) |

p P i

p P i

sim c p p i p

c i csim c p

Pure Collaborative Filtering: 1st . User Similarity

2nd. Rating Predictor

P(i): User similarity of c

c(i): Rating predicted for item i by c

p(i): Rating for item i by a producer p

sim(c, p):Similarity between c and p

Page 25: Trust and  Recommender Systems

25

Trust [N*N]

Rating [N*M]

Input

N: UsersM: Items

Trust Metric

EstimatedTrust[N*N]

SimilarityMetric

User [N*N]Similarity

Rating Predictor

PredictedRating[N*M]

OutputFirst step Second step

Pure Collaborative Filtering

Trust in Recommender System

Page 26: Trust and  Recommender Systems

26

From the Epinions.com Web site49,290 users who rated a total of139,738 different items at least once,

writing664,824 reviews.487,181 issued trust statements.

Consists of 2 filesRatings dataTrust data

Experimental Analysis

Dataset

Page 27: Trust and  Recommender Systems

Experimental Analysis

Page 28: Trust and  Recommender Systems

Experimental Analysis

Page 29: Trust and  Recommender Systems

Introduce Recommender System、MoleTrust、 TidalTrust、 PageRank

Trust is very effective in alleviating RSs weaknesses: Data sparsity; Be easily attacked; Cold-start.

Trust propogation is a tradeoff in terms of Accuracy and Coverage;

Conclusion

Page 30: Trust and  Recommender Systems

30

Thanks for your attention !