Elsevier

Applied Soft Computing

Volume 34, September 2015, Pages 693-704
Applied Soft Computing

GenTrust: A genetic trust management model for peer-to-peer systems

https://doi.org/10.1016/j.asoc.2015.04.053Get rights and content

Abstract

In recent years, peer-to-peer systems have attracted significant interest by offering diverse and easily accessible sharing environments to users. However, this flexibility of P2P systems introduces security vulnerabilities. Peers often interact with unknown or unfamiliar peers and become vulnerable to a wide variety of attacks. Therefore, having a robust trust management model is critical for such open environments in order to exclude unreliable peers from the system. In this study, a new trust model for peer-to-peer networks called GenTrust is proposed. GenTrust has evolved by using genetic programming. In this model, a peer calculates the trustworthiness of another peer based on the features extracted from past interactions and the recommendations. Since the proposed model does not rely on any central authority or global trust values, it suits the decentralized nature of P2P networks. Moreover, the experimental results show that the model is very effective against various attackers, namely individual, collaborative, and pseudospoofing attackers. An analysis on features is also carried out in order to explore their effects on the results. This is the first study which investigates the use of genetic programming on trust management.

Introduction

Open nature of peer-to-peer (P2P) systems facilitates join or leave of users to the network without worrying about any obligations. However this freedom creates potential threats for good behaving peers. Since there is no central authority to manage inter-peer interactions, malicious peers can easily perform attacks or take advantage of system resources without contributing to the system. A way to mitigate such threats is to create artificial trust relationships among users based on peer interactions. Trust models can help in such open environments to quantify trustworthiness numerically and create trust relationships among peers. However, it is hard to measure and formulate trust with numeric values. Furthermore, measuring trust without a priori knowledge is a challenging problem in P2P systems since peers mostly interact with unknown peers. Therefore, trust management in P2P environments is a difficult research problem.

When there is a central authority, trust management is relatively easy problem. In some e-commerce applications, a central authority collects user inputs about completed interactions and this information is used to make trust decisions about future interactions. Although fake users and interactions can pollute the collected information, this model mostly works on e-commerce applications. However, P2P systems need more complex trust management models due to the lack of a central authority. Peers need to store and manage trust information about each other [1], [2], [3]. On the other hand, uncertain information collected from neighboring peers might be deceptive. Malicious peers might deliberately provide wrong information to the system and this might not be detected since there is no central authority. Therefore, trust models in P2P systems should be able to recognize various attacks and help benign peers to find trustworthy peers. While doing this task, ambiguous information collected from other peers should be processed carefully to make correct decisions.

The trust decision problem can be considered as a classification problem, and machine learning techniques could be employed to distinguish malicious peers from benign peers. This paper proposes a genetic programming (GP) based trust management model (GenTrust), extending our previous work [4] with greater experimental verification and analysis on features. The proposed model helps to identify malicious peers and find trustworthy peers using the features derived from peer interactions and recommendations. The model has evolved with these features by using genetic programming, which provides a mathematical function to measure trust values of peers. A peer ranks its neighbors according to trust values, and makes trusting decisions using these values. Each peer stores trust relationships for the peers they have interacted in the past. As peers gain more neighbors with time, malicious peers are excluded from the system using trust relationships. The evolved model is evaluated against various attackers, namely individual attackers, collaborators, and pseudospoofers. The results show that the model decreases the number of attacks considerably. Features of the model are also analyzed and their effects on the performance is assessed. Satisfaction related features are found more influential in trust decisions. Cross training and testing are performed among various attacker types to understand the model's adaptability on different attacker behaviors. These experiments show that the models trained on complex attack behaviors are also successful in simple attack behaviors.

Organization of the paper is as follows. Section 2 gives a summary of the state of the art research. Sections 3 and 4 explain the proposed trust model and the simulation environment respectively. Section 5 presents the experiments and discusses their results extensively. Section 6 outlines the conclusions of the study.

Section snippets

Related work

Trust is a social concept and hard to measure and formalize with theoretical foundations. Although some approaches have formulated trust as a result of direct experiences [5], most trust models use recommendations of others to build trust relationships [6], [7], [8]. However, it might be hard to correctly evaluate trustworthiness using recommendations, since recommendations may contain deceptive or subjective opinions [9]. To better address different aspects of the trust, some approaches use

The model

Measuring trust without a priori knowledge is a challenging problem in P2P systems since peers mostly interact with unknown peers. In this research, GP is proposed to discover automatically complex properties of P2P networks. GP is a common evolutionary computation technique introduced by Koza [59]. Evolutionary computation (EC) is one of the most promising approaches in intrusion detection. A recent survey [49] showed that evolutionary computation techniques allow to obtain more readable

The attacks

In this section, the attack models considered in the study are outlined. Defining representative and realistic attack model is very important to evaluate a trust model successfully. The attacks covered here range from simple attacks to hard attacks in terms of detection. Attacks taking advantage of evasion strategies such as collaboration and pseudospoofing are also taken into account in the model.

In a P2P network, benign peers always behave as expected and properly carry out their tasks such

Analysis on individual attackers

First of all, a trust model is generated for individual attackers. In the training, the generated individuals are simulated on a network in which 10% of the peers is malicious. The GP algorithm is run ten times for each attack type and the best results is evaluated in testing. Testing is done on networks with different amounts of malicious peers (10%, 30% and 50%). In the experiments, the attack probability of hypocritical attackers is set 20% for all interactions.

Table 3 shows the success

Conclusion

This paper proposes GenTrust, a trust model evolved using genetic programming. The generated model allows each peer to calculate trust values of other peers based on interaction and recommendation based features. Naive and hypocritical attacker models are studied with individual, collaborative, and pseudospoofing behaviors. The model is trained against these attack types and evaluated on various network setups containing different ratio of malicious peers. The experimental results show that the

References (65)

  • S. Marsh

    Formalising trust as a computational concept

    (1994)
  • A. Abdul-Rahman et al.

    Supporting trust in virtual communities

  • B. Yu et al.

    A social mechanism of reputation management in electronic communities

  • L. Mui et al.

    A computational model of trust and reputation for e-businesses

  • A. Jøsang et al.

    Analysing topologies of transitive trust

  • R. Guha et al.

    Propagation of trust and distrust

  • D. Gefen

    Reflections on the dimensions of trust and trustworthiness among online consumers

    SIGMIS Database

    (2002)
  • G. Swamynathan et al.

    Decoupling service and feedback trust in a peer-to-peer reputation system.

  • P. Resnick et al.

    Reputation systems

    Commun. ACM

    (2000)
  • C. Dellarocas

    Immunizing online reputation reporting systems against unfair ratings and discriminatory behavior

  • J. Douceur

    The sybil attack

  • H. Yu et al.

    Sybilguard: defending against sybil attacks via social networks

    SIGCOMM Comput. Commun. Rev.

    (2006)
  • N. Tran et al.

    Sybil-resilient online content voting

  • E.J. Friedman et al.

    The social cost of cheap pseudonyms

    J. Econ. Manag. Strateg.

    (2001)
  • Z. Despotovic et al.

    Trust-aware delivery of composite goods

  • T. Bearly et al.

    Building trust and security in peer-to-peer systems

  • L. Mekouar et al.

    Reputation-based trust management in peer-to-peer systems: taxonomy and anatomy

  • N. Stakhanova et al.

    A reputation-based trust management in peer-to-peer network systems

  • A.A. Selcuk et al.

    A reputation-based trust management system for p2p networks

  • R. Zhou et al.

    Gossiptrust for fast reputation aggregation in peer-to-peer networks

    IEEE Trans. Knowl. Data Eng.

    (2008)
  • L. Xiong et al.

    Peertrust Supporting reputation-based trust for peer-to-peer ecommerce communities

    IEEE Trans. Knowl. Data Eng.

    (2004)
  • I. Stoica et al.

    Chord: A scalable peer-to-peer lookup service for internet applications

    ACM SIGCOMM Comput. Commun. Rev.

    (2001)
  • Cited by (0)

    This paper is an extended, improved version of the paper “Evolving a Trust Model for Peer-To-Peer Networks Using Genetic Programming” presented at EvoComNet2014 and published in: Applications of Evolutionary Computing, Proceedings of 17th European Conference, EvoApplications 2014, Granada, Spain, April 23–25, 2014, LNCS 8602, Springer, 2014.

    View full text