• Aucun résultat trouvé

TRUTHFUL REPUTATION MECHANISMS FOR ONLINE SYSTEMS

N/A
N/A
Protected

Academic year: 2022

Partager "TRUTHFUL REPUTATION MECHANISMS FOR ONLINE SYSTEMS"

Copied!
7
0
0

Texte intégral

(1)

POUR L'OBTENTION DU GRADE DE DOCTEUR ÈS SCIENCES

PAR

ingénieur en informatique diplômé de l'Université Polytechnique de Timisoara, Roumanie et de nationalité roumaine

acceptée sur proposition du jury:

Lausanne, EPFL 2007

Prof. T. Henzinger, président du jury Prof. B. Faltings, directeur de thèse

Prof. K. Aberer, rapporteur Prof. C. Dellarocas, rapporteur

Prof. T. Sandholm, rapporteur

TRUTHFUL REPUTATION MECHANISMS FOR ONLINE SYSTEMS

Radu JURCA

THÈSE N

O

3955 (2007)

ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE

PRÉSENTÉE LE 29 OCTOBRE 2007

À LA FACULTÉ INFORMATIQUE ET COMMUNICATIONS Laboratoire d'intelligence artifi cielle

SECTION D'INFORMATIQUE

(2)

iii

R´ esum´ e

L’internet constitue aujourd’hui un milieu interactif o`u les communaut´es et les ´economies virtuelles gagnent de l’importance par rapport `a leurs contreparties traditionnelles. Tandis que ce d´ecalage cr´ee des occasions et avantages qui ont d´ej`a am´elior´es notre vie quotidienne, il apporte ´egalement une toute nouvelle s´erie de probl`emes. Par exemple, le manque d’interaction physique qui caract´erise la majorit´e des transactions ´electroniques rend les syst`emes beaucoup plus susceptibles `a la fraude et `a la tromperie.

Les m´ecanismes de r´eputation offrent une mani`ere nouvelle et efficace d’assurer la confiance qui est essentielle au fonctionnement de chaque march´e. Ils rassemblent les informations sur l’histoire (c.-`a-d.

les transactions ant´erieures) des agents qui participent dans le march´e, et publient leur r´eputation.

Les futurs associ´es guident leurs d´ecisions en consid´erant l’information sur la r´eputation, et sont ainsi capable de faire les meilleurs choix. Les m´ecanismes de r´eputation en ligne connaissent un succ`es remarquable: ils sont pr´esents dans la plupart des syst`emes commerciaux d´eploy´es aujourd’hui, et sont s´erieusement consult´es par les utilisateurs humains.

La valeur ´economique de la r´eputation en ligne soul`eve des questions concernant la fiabilit´e des m´ecanismes eux-mˆemes. Les syst`emes actuels ont ´et´e con¸cus en supposant que les utilisateurs partageront honnˆetement leurs avis. Cependant, des ´etudes r´ecentes ont d´emontr´e qu’il existe des utilisateurs qui d´enaturent la v´erit´e pour manipuler la r´eputation.

La pr´esente th`ese d´ecrit diff´erentes mani`eres de rendre les m´ecanismes de r´eputation en ligne plus dignes de confiance, en encouragent les participants `a communiquer honnˆetement les informations qu’ils d´etiennent. Diff´erents types de m´ecanismes de r´eputation sont ´etudi´es, et pour chacun, des m´ecanismes pour r´ecompenser les agents qui rapportent la v´erit´e sont pr´esent´es. Les probl`emes li´es `a la complicit´e (c.-`a-d. la coordination de la strat´egie de plusieurs agents afin de manipuler le syst`eme) et `a la ro- bustesse sont ´egalement ´etudi´es. De plus, cette th`ese d´ecrit une nouvelle application des m´ecanismes de r´eputation pour surveiller la qualit´e livr´ee par des fournisseurs de services, et ´etudie les facteurs qui motivent et influencent des utilisateurs humains qui postent leurs avis dans des forums existants.

Mots cl´es: m´ecanismes de r´eputation en ligne, feedback, incentive-compatibility, la collusion, mechanism design, la th´eorie des jeux

(3)

v

Abstract

The internet is moving rapidly towards an interactive milieu where online communities and economies gain importance over their traditional counterparts. While this shift creates opportunities and benefits that have already improved our day-to-day life, it also brings a whole new set of problems. For example, the lack of physical interaction that characterizes most electronic transactions, leaves the systems much more susceptible to fraud and deception.

Reputation mechanisms offer a novel and effective way of ensuring the necessary level of trust which is essential to the functioning of any market. They collect information about the history (i.e., past transactions) of market participants and make public theirreputation. Prospective partners guide their decisions by considering reputation information, and thus make more informative choices. Online reputation mechanisms enjoy huge success. They are present in most e-commerce sites available today, and are seriously taken into consideration by human users.

The economical value of online reputation raises questions regarding the trustworthiness of mecha- nisms themselves. Existing systems were conceived with the assumption that users will share feedback honestly. However, we have recently seen increasing evidence that some users strategically manipulate their reports.

This thesis describes ways of making online reputation mechanisms more trustworthy by providing incentives to rational agents for reporting honest feedback. Different kinds of reputation mechanisms are investigated, and for each, I present mechanisms for rewarding the agents that report truthfully.

Problems related to collusion (i.e., several agents coordinate their strategies in order to manipulate reputation information) and robustness are also investigated. Moreover, this thesis describes a novel application of incentive compatible reputation mechanisms to the area of quality of service monitoring, and investigates factors that motivate and bias human users when reporting feedback in existing review forums.

Keywords: Online reputation mechanisms, feedback, incentive-compatibility, collusion, reporting behavior, mechanism design, game theory

(4)

Contents

1 Introduction 1

1.1 Summary of the Thesis . . . 5

2 Trust, Reputation and Reputation Mechanisms 9 2.1 Modelling Trust . . . 10

2.2 Reputation . . . 12

2.2.1 The nature of reputation information . . . 13

2.2.2 The role of reputation information . . . 14

2.3 Reputation Mechanisms for Online Systems . . . 25

2.3.1 Online Implementations . . . 26

2.3.2 Academic Models . . . 28

2.3.3 Empirical studies of Reputation Mechanisms . . . 30

2.3.4 Other Aspects related to Reputation Mechanisms . . . 32

3 Truthful Signaling Reputation Mechanisms 33 3.1 A Formal Model . . . 34

3.2 Incentives for Honestly Reporting Feedback . . . 36

3.2.1 Incentive-compatible Payment Mechanisms . . . 37

3.3 Automated Design of Incentive-compatible Payment Mechanisms . . . 38

3.3.1 Example . . . 40

3.3.2 Unknown Lying Incentives . . . 42

3.3.3 Computational Complexity and Possible Approximations . . . 42

3.4 Further Decreasing the Feedback Payments . . . 45

3.4.1 Using Several Reference Reports . . . 45

ix

(5)

x Contents

3.4.2 Filtering out False Reports . . . 48

3.5 Robust Incentive-Compatible Payment Mechanisms . . . 52

3.5.1 Dishonest Reporting with Unknown Beliefs . . . 52

3.5.2 Declaration of Private Information . . . 53

3.5.3 Computing Robust Incentive-Compatible Payments . . . 55

3.5.4 General Tolerance Intervals for Private Information . . . 56

3.6 Collusion-resistant, Incentive-compatible Rewards . . . 58

3.6.1 Collusion Opportunities in Binary Payment Mechanisms . . . 59

3.6.2 Automated Design of Collusion-resistant Payments . . . 63

3.6.3 Full Coalitions on Symmetric Strategies, Non-Transferable Utilities . . . 65

3.6.4 Full Coalitions on Asymmetric Strategies, Non-Transferable Utilities . . . 68

3.6.5 Partial Coalitions on Symmetric Strategies, Non-Transferable Utilities . . . 72

3.6.6 Partial Coalitions on Asymmetric Strategies, Non-Transferable Utilities . . . 74

3.6.7 Partial Coalitions on Asymmetric Strategies, Transferable Utilities . . . 79

3.7 Related Work . . . 81

3.8 Summary of Results . . . 84

3.A Summary of Notation . . . 87

3.B Generating Random Settings . . . 88

3.C Cardinality ofQ(Nref) . . . 88

3.D Proof of Lemma 3.6.1 . . . 89

3.E Generating Random Binary Settings . . . 89

3.F Proof of Proposition 3.6.7 . . . 90

3.G Proof of Proposition 3.6.8 . . . 91

4 Novel Applications of Signaling Reputation Mechanisms - QoS Monitoring 93 4.1 Formal Model and Assumptions . . . 96

4.2 Interaction Protocol . . . 98

4.3 Implementation of a Prototype . . . 99

4.4 Incentive-compatible Service Level Agreements . . . 101

4.4.1 Example of Incentive-compatible SLAs . . . 102

4.5 Reliable QoS Monitoring . . . 104

(6)

Contents xi

4.5.1 Example . . . 108

4.6 Deterring Malicious Coalitions . . . 108

4.6.1 Using Trusted Monitoring Infrastructure . . . 110

4.7 Summary of Results . . . 112

4.A Summary of Notation . . . 113

5 Sanctioning Reputation Mechanisms 115 5.1 Related Work . . . 118

5.2 The Setting . . . 119

5.2.1 Example . . . 120

5.2.2 Strategies and Equilibria . . . 121

5.2.3 Efficient Equilibrium Strategies . . . 125

5.3 Designing Efficient Reputation Mechanisms . . . 127

5.3.1 Probabilistic Reputation Mechanisms . . . 128

5.3.2 Deterministic Reputation Mechanisms using Mixed Strategies . . . 130

5.3.3 Deterministic Reputation Mechanisms with Pure Strategies . . . 135

5.3.4 Feedback Granularity . . . 137

5.4 A Mechanism for Obtaining Reliable Feedback Reports . . . 138

5.4.1 The CONFESS Mechanism . . . 141

5.4.2 Behavior and Reporting Incentives . . . 142

5.4.3 Implementation in the Reputation Mechanism . . . 144

5.4.4 Analysis of Equilibria . . . 145

5.4.5 Building a Reputation for Truthful Reporting . . . 149

5.4.6 The Threat of Malicious Clients . . . 154

5.4.7 Remarks . . . 155

5.5 Summary of Results . . . 157

5.A Summary of Notation . . . 159

5.B Appendix: Proof of Proposition 5.2.1 . . . 159

5.C Appendix: Proof of Proposition 5.2.2 . . . 160

5.D Proof of Proposition 5.4.2 . . . 161

5.E Proof of Proposition 5.4.3 . . . 163

(7)

xii Contents

6 Understanding Existing Online Feedback 165

6.1 The Data Set . . . 166

6.1.1 Formal notation . . . 168

6.2 Evidence from Textual Comments . . . 168

6.2.1 Correlation between Reporting Effort and Transactional Risk . . . 174

6.3 The Influence of Past Ratings . . . 178

6.3.1 Prior Expectations . . . 179

6.3.2 Impact of Textual Comments on Quality Expectation . . . 180

6.3.3 Reporting Incentives . . . 181

6.4 Modelling the Behavior of Raters . . . 182

6.4.1 Model Validation . . . 183

6.5 Summary of Results . . . 185

6.A List of words,LR, associated to the feature Rooms . . . 186

7 Conclusions 187 7.1 Directions for Future Work . . . 190

7.1.1 From “lists of reviews” to designed reputation mechanisms . . . 190

7.1.2 Signaling and sanctioning reputation mechanisms . . . 191

7.1.3 Factoring human behavior into reputation mechanism design . . . 191

7.1.4 Mechanisms for social networks and P2P systems . . . 192

7.1.5 Reputation mechanisms translated to other domains . . . 192

Bibliography 193

Références

Documents relatifs

Lab 2014 developed an evaluation methodology and test collections for two dif- ferent reputation management problems: (1) classification of tweets according to the

In the talk we will summarize some of the main challenges that Information Access Technologies must face to assist online reputa- tion monitoring tasks, and present some of the

Considered together with the poor prognosis of patients with primary oxaluria on any form of dialysis therapy, these limited but rather optimistic reports probably encouraged

Confocal scanning laser microscopy has been combined with reporter gene technology to visualise bacterial signal molecules (N-acyl-homoserine lactones) produced on plant roots and

We design an incentive mechanism under which the content provider rewards the nodes based on the data transmission rate they contribute.. The data transmission rate includes the

On the other hand, whenever a requesting peer has to face a large proportion of malicious peers, it can only rely on its own feedback to estimate the effort exerted by the target

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

Upon receipt of "enough" feedback, the requesting peer aggregates them with its own observations (if any) to estimate the reputation of the target server, and provides