SpamResist: Making Peer-to-Peer Tagging SpamResist: Making - - PowerPoint PPT Presentation

▶
spamresist making peer to peer tagging spamresist making
SMART_READER_LITE
LIVE PREVIEW

SpamResist: Making Peer-to-Peer Tagging SpamResist: Making - - PowerPoint PPT Presentation

Dec 15, 2022 8 likes •656 views

SpamResist: Making Peer-to-Peer Tagging SpamResist: Making Peer-to-Peer Tagging Systems Robust to Spam Systems Robust to Spam Ennan Ennan Zhai, Zhai, Ruichuan Ruichuan Chen, Eng Keong hen, Eng Keong Lua* Lua* Long Zhang, Long Zhang,

slide-1
SLIDE 1

SpamResist: Making Peer-to-Peer Tagging SpamResist: Making Peer-to-Peer Tagging Systems Robust to Spam Systems Robust to Spam

Ennan Ennan Zhai, Zhai, Ruichuan Ruichuan Chen, Eng Keong hen, Eng Keong Lua* Lua* Long Zhang, Long Zhang, Huiping Huiping Sun, Zhuhua un, Zhuhua Cai Cai Sihan Sihan Qing, Zhong ing, Zhong Chen Chen Peking University & *Carnegie Mellon University Peking University & *Carnegie Mellon University

slide-2
SLIDE 2

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

1

slide-3
SLIDE 3

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

1

slide-4
SLIDE 4

There are some tagging services-based systems in our lives … …

Tagging Systems Tagging Systems

2

slide-5
SLIDE 5

To meet the challenge, such as DoS or single point failure, tagging services are introduced into P2P content systems… …

P2P Tagging Systems P2P Tagging Systems

3

slide-6
SLIDE 6

To meet the challenge, such as DoS or single point failure, tagging services are introduced into P2P content systems… …

P2P Tagging Systems P2P Tagging Systems

For example, Tagster is an open source DHT-based P2P tagging system.

(http://isweb.uni-koblenz.de/research/tagster)

3

slide-7
SLIDE 7

The results for searching tag “iphone” in MyWeb.

Tag Spam Tag Spam

4

slide-8
SLIDE 8

When we click this link, we will find the following page …….

Tag Spam Tag Spam

The results for searching tag “iphone” in MyWeb.

4

slide-9
SLIDE 9

This Figure is not related to iphones.

Tag Spam Tag Spam

5

slide-10
SLIDE 10

This Figure is not related to iphones.

Tag Spam Tag Spam

5

We can also observe that this site has been assigned many other popular but irrelevant tags.

slide-11
SLIDE 11

That is the problem of tag spam! Definition of Tag Spam: The erroneous or misleading tags that are generated by some malicious users to confuse the normal users in the systems.

Tag Spam Tag Spam

6

slide-12
SLIDE 12

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

7

slide-13
SLIDE 13

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

7

slide-14
SLIDE 14

Detection-based Mechanisms. Demotion-based Mechanisms. Interface-based Mechanisms.

Related Work Related Work

8

slide-15
SLIDE 15

Detection-based Mechanisms. Demotion-based Mechanisms. Interface-based Mechanisms.

Related Work Related Work

8

slide-16
SLIDE 16

Detection-based Mechanisms. Demotion-based Mechanisms. Interface-based Mechanisms.

Related Work Related Work

8

slide-17
SLIDE 17

Detection-based Mechanisms. Demotion-based Mechanisms. Interface-based Mechanisms.

Related Work Related Work

8

slide-18
SLIDE 18

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

9

slide-19
SLIDE 19

What is the tag spam in P2P tagging What is the tag spam in P2P tagging systems? systems? What are the existing solutions on this What are the existing solutions on this problem? problem? Our approach? Our approach?

I I I I I I

Roadmap Roadmap

9

slide-20
SLIDE 20

SpamResist is a demotion-based mechanism, and encompasses two key parts: Reliability Mechanism; Social Network-based Enhancement.

SpamResist SpamResist

10

slide-21
SLIDE 21

SpamResist is a demotion-based mechanism, and encompasses two key parts: Reliability Mechanism; Social Network-based Enhancement.

SpamResist SpamResist

10

slide-22
SLIDE 22

For each tag (e.g., Sea) search, client calculates a reliability degree for each respondent, and uses weighted averaging to compute the rank of the search results.

What is reliability mechanism? What is reliability mechanism?

11

slide-23
SLIDE 23

For each tag (e.g., Sea) search, client calculates a reliability degree for each respondent, and uses weighted averaging to compute the rank of the search results. The peer who annotated some local files with “Sea” will respond the client with these files. We call this peer as respondent.

What is reliability mechanism? What is reliability mechanism?

11

slide-24
SLIDE 24

For each tag (e.g., Sea) search, client calculates a reliability degree for each respondent, and uses weighted averaging to compute the rank of the search results.

What is reliability mechanism? What is reliability mechanism?

Weight is the reliability degree of the owner of each response resource.

11

slide-25
SLIDE 25

Weight is the reliability degree of the owner of each response resource. For each tag (e.g., Sea) search, client calculates a reliability degree for each respondent, and uses weighted averaging to compute the rank of the search results.

What is reliability mechanism? What is reliability mechanism?

How the client to compute the reliability degree for each peer? 11

slide-26
SLIDE 26

Reliability degree is a personalized score assigned to each peer by the client, and SpamResist proposes two schemes for the client to calculate the reliability degrees of two categories of peers respectively:

  • Unfamiliar peers;
  • Interacted peers.

How to compute reliability? How to compute reliability?

12

slide-27
SLIDE 27

Reliability degree is a personalized score assigned to each peer by the client, and SpamResist proposes two schemes for the client to calculate the reliability degrees of two categories of peers respectively:

  • Unfamiliar peers;
  • Interacted peers.

How to compute reliability? How to compute reliability?

Normally, the behaviors that peer A downloads some files from peer B are called interactions between A and B.

12

slide-28
SLIDE 28

Reliability degree is a personalized score assigned to each peer by the client, and SpamResist proposes two schemes for the client to calculate the reliability degrees of two categories of peers respectively:

  • Unfamiliar peers;
  • Interacted peers.

How to compute reliability? How to compute reliability?

The peers that the client has never interacted with.

12

slide-29
SLIDE 29

Unfamiliar Peer’s Reliability Unfamiliar Peer’s Reliability

13

slide-30
SLIDE 30

Reliability degree is a personalized score assigned to each peer by the client, and SpamResist proposes two schemes for the client to calculate the reliability degrees of two categories of peers respectively:

  • Unfamiliar peers;
  • Interacted peers.

How to compute reliability? How to compute reliability?

The peers that the client has interacted with.

14

slide-31
SLIDE 31

Interacted Peer’s Reliability Interacted Peer’s Reliability

The client stores the previous experiences from the interacted peers in his own experience vector (EVA,B).

n B A B A B A

v v v

, , 2 , , 1 , ,

,..., ,

15

slide-32
SLIDE 32

The client stores the previous experiences from the interacted peers in his own experience vector (EVA,B).

Interacted Peer’s Reliability Interacted Peer’s Reliability

n B A B A B A

v v v

, , 2 , , 1 , ,

,..., ,

Specifically, for the peer B that client A has interacted with, A maintains a vector of length n storing the most recent n experiences with B, and as new experiences are appended the oldest ones are removed.

15

slide-33
SLIDE 33

Interacted Peer’s Reliability Interacted Peer’s Reliability

16

slide-34
SLIDE 34

The client stores the previous experiences from the interacted peers in his own experience vector (EVA,B).

Interacted Peer’s Reliability Interacted Peer’s Reliability

n B A B A B A

v v v

, , 2 , , 1 , ,

,..., ,

Reliability degree from A to B (interacted peer for A) is:

n v v v

n B A B A B A , , 2 , , 1 , ,

...+ + +

17

slide-35
SLIDE 35

SpamResist is a demotion-based mechanism, and encompasses two key parts: Reliability Mechanism; Social Network-based Enhancement.

SpamResist SpamResist

18

slide-36
SLIDE 36

SpamResist is a demotion-based mechanism, and encompasses two key parts: Reliability Mechanism; Social Network-based Enhancement.

SpamResist SpamResist

18

slide-37
SLIDE 37

Social Network-based Enhancement Social Network-based Enhancement

  • Re-compute the ranking score (RS) for the result file

whose RS is lower than 0.5.

  • If more than half of friends have RS higher than 0.5,

re-locate the position.

  • According to average of scores higher than 0.5.

19

slide-38
SLIDE 38

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 Alice

20

slide-39
SLIDE 39

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 Alice Alice’s Friends

20

slide-40
SLIDE 40

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 0.7 0.8 0.4 Ranking Scores Alice Alice’s Friends

20

slide-41
SLIDE 41

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 (0.7 + 0.8) / 2 = 0.75 Alice Alice’s Friends

20

slide-42
SLIDE 42

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 (0.7 + 0.8) / 2 = 0.75 Alice Alice’s Friends

20

slide-43
SLIDE 43

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.75 (0.7 + 0.8) / 2 = 0.75 Alice Alice’s Friends

20

slide-44
SLIDE 44

Social Network-based Enhancement Social Network-based Enhancement

sea.jpg Alice’s search result … … Ranking Score … … 0.4 0.5 0.3 0.2 Ranking Score Alice Alice’s Friends

20

slide-45
SLIDE 45
  • The details about social network-based enhancement

mechanism of SpamResist please see our paper.

  • The practical issue on unreliable friends please see
  • ur paper.

Social Network-based Enhancement Social Network-based Enhancement

21

slide-46
SLIDE 46
  • The details about social network-based enhancement

mechanism of SpamResist please see our paper.

  • The practical issue on unreliable friends please see
  • ur paper.

Social Network-based Enhancement Social Network-based Enhancement

21

slide-47
SLIDE 47
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

22

slide-48
SLIDE 48
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

22

slide-49
SLIDE 49

The search strategy of Boolean is the system randomly ranks the results associated with the search tag.

Boolean Model Boolean Model

23

slide-50
SLIDE 50
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

24

slide-51
SLIDE 51
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

24

slide-52
SLIDE 52

Occurrence model ranks the search results based on the number of annotations containing the searched tag and returns the top ranking results.

Occurrence Model Occurrence Model

25

slide-53
SLIDE 53
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

26

slide-54
SLIDE 54
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

26

slide-55
SLIDE 55

Coincidence model assigns each user a score computed by the number of the annotations overlapped with other users in the system.

Coincidence Model Coincidence Model

27

slide-56
SLIDE 56
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

28

slide-57
SLIDE 57
  • Boolean Model.
  • Occurrence Model.
  • Coincidence Model.
  • PINTS.

Search Models:

Evaluation Evaluation

28

slide-58
SLIDE 58

The strategy of PINTS is the client first generates a feature vector to store a characteristic score for each peer in system; then, using the vector, the client aggregates the information of annotations selectively, and randomly ranks the search result.

PINTS PINTS

29

slide-59
SLIDE 59
  • Random Attacks: randomly annotate misleading tags

to the resources in the system;

  • Targeted Attacks: collusively annotate resources with

the same misleading tags;

  • Tricky Attacks: annotate resources with both correct

and misleading tags. This attack could make some anti- spam scheme unusable.

Evaluation Evaluation

Threat Models: 30

slide-60
SLIDE 60

The impact of random attack under 20% random attackers (more detail see paper)

Evaluation Evaluation

31

slide-61
SLIDE 61

Evaluation Evaluation

The impact of targeted attack under 20% targeted attackers (more detail see paper)

32

slide-62
SLIDE 62

Evaluation Evaluation

The impact of tricky attack under 20% tricky attackers (more detail see paper)

33

slide-63
SLIDE 63

SpamResist is a novel social reliability-based mechanism towards spam-free and personalized tag search results in the P2P tagging systems.

Conclusion Conclusion

34

slide-64
SLIDE 64

Q & A