Guardian : Evaluating Trust in Online Social Networks with Graph - - PowerPoint PPT Presentation

▶
guardian evaluating trust in online social networks with
SMART_READER_LITE
LIVE PREVIEW

Guardian : Evaluating Trust in Online Social Networks with Graph - - PowerPoint PPT Presentation

Aug 20, 2023 369 likes •737 views

Guardian : Evaluating Trust in Online Social Networks with Graph Convolutional Networks Wanyu Lin , Zhaolin Gao, Baochun Li University of Toronto Almost 4.57 billion people were active internet users as of April 2020. Statista Social trust

slide-1
SLIDE 1

Wanyu Lin, Zhaolin Gao, Baochun Li University of Toronto

Guardian: Evaluating Trust in Online Social Networks with Graph Convolutional Networks

slide-2
SLIDE 2

Almost 4.57 billion people were active internet users as of April 2020. — Statista

slide-3
SLIDE 3

Social trust is the basis of

  • nline social networks.
slide-4
SLIDE 4

Estimates of social trust help indicate to what extent a user could expect someone else to perform given actions, therefore has many applications, such as trust-based recommendations.

slide-5
SLIDE 5

Network graph

an example

H G D A B E C F

2 1 1 1 3 3 1 3 2

slide-6
SLIDE 6

H G D A B E C F

2 1 1 1 3 3 1 3 2

  • I-
  • i
  • Can A trust E? And, to what extent?
slide-7
SLIDE 7

Large-scale

slide-8
SLIDE 8

Wait a second …

slide-9
SLIDE 9

Graph convolutional neural networks — an efficient variant

  • f convolutional neural networks
  • n graphs.

9

slide-10
SLIDE 10

10

Grungier

GrghGni_n.FI

Output

  • i

if

  • lilli
  • __

__

ei

__

Representation learning with graph convolutional networks

slide-11
SLIDE 11

The complexity of model parameters are independent of the input graph size.

slide-12
SLIDE 12

Preliminaries: trust properties

H G D A B E C F

2 1 1 1 3 3 1 3 2

slide-13
SLIDE 13

Trust properties

G E

1 3

H G D A B E C F

2 1 1 1 3 3 1 3 2

Asymmetry: one user may trust someone else more than she is trusted back.

slide-14
SLIDE 14

Trust properties

A B C

3 2 ?

H G D A B E C F

2 1 1 1 3 3 1 3 2

Propagative nature: trust may be passed from

  • ne user to another, creating chains of social

trust that connects two users who are not connected.

slide-15
SLIDE 15

Trust properties

H G D A B E C F

2 1 1 1 3 3 1 3 2

Composable nature: trust needs to be aggregated if several chains of social trust exit.

D A B E

? 1 3 3 1

slide-16
SLIDE 16

An effective way of evaluating trust should be able to capture these trust properties simultaneously.

16

slide-17
SLIDE 17

Guardian: an end-to-end learning framework for social trust evaluation.

17

slide-18
SLIDE 18

H G D A B E C F

2 1 1 1 3 3 1 3 2

I

z

I

3

  • I
  • s

z

I

  • I

3

B
  • 8

O

E

z

I

1

±

: g

Z

I

I

3

B
  • g

I

O

z

I

3

I

  • I
  • ,
  • I

I

3

B
  • g

I

I

z

I

3

B
  • y
8

1

I

z

I

3

B
  • g

I

Embedding layer

We use a pre-trained embedding layer that maps each user into a vector.

18

slide-19
SLIDE 19

Two types of trust interactions: popularity trust and engagement trust

19

slide-20
SLIDE 20

Popularity trust: the overall trust of a user endorsed by others (accumulated from the incoming links)

H G D A B E C F

2 1 1 1 3 3 1 3 2

  • I-
  • i
slide-21
SLIDE 21

H G D A B E C F

2 1 1 1 3 3 1 3 2

  • I-
  • i
  • Engagement trust: the willingness of a

user to trust others (accumulated from the outgoing links)

slide-22
SLIDE 22

H G D A B E C F

2 1 1 1 3 3 1 3 2

  • I-
  • i
  • Two types of trust

aggregation

D A B

3 3

A F

2

Popularity Engagement

G E

3

H G D B E

1 1 1 1

slide-23
SLIDE 23

23

Embedding Layer

I 0

Engagement Trust Propagation

I 1

... ...

I i O O

n

O

1

Trust Convolutional Layers

Concatenation ?

Prediction Layer

Trust Relationship

Fully Connected Layer

Softmax Popularity Trust Propagation ? Trustor Trustee Trustor Trustee Mean Mean

Trust convolutional layer

To capture the composable and asymmetric nature of trust

slide-24
SLIDE 24

Embedding Layer

I 0

Engagement Trust Propagation

I 1

... ...

I i O O

n

O

1

Trust Convolutional Layers

Concatenation ?

Prediction Layer

Trust Relationship

Fully Connected Layer

Softmax Popularity Trust Propagation ? Trustor Trustee Trustor Trustee Mean Mean

Stack multiple trust convolutional layers

To capture the propagative nature of trust

slide-25
SLIDE 25

Prediction layer

25

Embedding Layer

I 0

Engagement Trust Propagation

I 1

... ...

I i O O

n

O

1

Trust Convolutional Layers

Concatenation ?

Prediction Layer

Trust Relationship

Fully Connected Layer

Softmax Popularity Trust Propagation ? Trustor Trustee Trustor Trustee Mean Mean

slide-26
SLIDE 26

Guardian

Embedding Layer

I 0

Engagement Trust Propagation

I 1

... ...

I i O O

n

O

1

Trust Convolutional Layers

Concatenation ?

Prediction Layer

Trust Relationship

Fully Connected Layer

Softmax Popularity Trust Propagation ? Trustor Trustee Trustor Trustee Mean Mean

slide-27
SLIDE 27

Our experimental results…

27

slide-28
SLIDE 28

Datasets Used

28

Advogato and Pretty-Good-Privacy (PGP) adopt the concept

  • f the “web of trust”, and both contain four different levels of

trust.

slide-29
SLIDE 29

Accuracy

29

Evaluation Accuracy on Advogato

slide-30
SLIDE 30

Accuracy

30

Evaluation Accuracy on PGP

slide-31
SLIDE 31

Efficiency

31

slide-32
SLIDE 32

Scalability

32

Time vs. # of pairs

slide-33
SLIDE 33

Scalability

33

Time vs. # of users

slide-34
SLIDE 34

Guardian is an end-to-end learning framework, that can achieve the best possible performance for social trust evaluation in online social networks.

slide-35
SLIDE 35

Wanyu Lin, Zhaolin Gao, Baochun Li University of Toronto

35