Computational Models Of Trust Restoration Framework In Multi-Agent Systems

The emergence of reputation is primarily a result of the occurrence of an incomplete knowledge on a specific issue and a full investigation and evaluation which is beyond the resource of an individual.The concept of reputation then comes into play by relying on aggregated power from multiple information sources.In multi-agent systems,reputation from its participants is particularly a central concern of many long-lived agents.Generally speaking,reputation plays a key role to improve cooperation among agents in its society by providing a decision-making tool for understanding the behavior of interacting partners.In fact,the success of agents' cooperation crucially depends on trust which other agents share with each other.However,among its great convenience due to the open,dynamic,and anonymous nature of multi-agent systems,cooperation in agent society has still encountered many challenges.One of the major challenges is multi-agent systems such as online environments are characterized as the environments with the present of noise.As a consequence,errors or mistakes are understood to be easily happened,pose problems for the success of a cooperative strategy in such systems.Errors or mistakes do not only cause the actual outcomes of agents'transactions to be worse than their expectations,but also cause the violation of trust to agents who delivered unexpected results,providing lowered tendency of future interactions.Broadly,trust violation during cooperation of autonomous agents in multi-agent systems is usually unavoidable and can arise due to a wide number of reasons.In the context of trust-based agent systems,trust is developed through positive experiences,meanwhile,negative experiences can violate trust at any moment during interaction.From a psychological point of view,the violation of an agent's trust is a result of one agent(which is a transgressor)expressing a very low weight on the welfare of another agent(which is a victim)by inflicting a high cost for a very small benefit.Moreover,the perception of low trustworthiness as a result of trust violation by an individual agent can be generalized to the whole agent society.In line with this,the violation of trust as a result of interactions that do not proceed as expected gives rise to the question as to whether broken trust can possibly be restored.However,the challenge of trust restoration is not to devise the mechanism to prevent a transgressor from violating trust again.But it is rather to devise the mechanism to evaluate the possibility a victim can regain a transgressor's trust and revive the expectations of good intentionsClearly,trust restoration is more complex than trust initialization and maintenance.In previous works,many approaches based on both non-psychological and psychological approaches have been introduced to deal with the process of trust restoration.However,the major distinct between their approaches and our approaches is most of them consider applying only one single aspect such as regret,apology or forgetting based on a quiescence time,to evaluate the possibility to forgive a transgressor.Therefore,trust restoration requires a more complex mechanism to explore different factors that cause the decline of trust and identify the affected individuals of trust violation both directly and indirectly.In this thesis,the following novel works are contributed:(1)Computational model of forgiveness mechanism is introduced to explore potential untrustworthy agents who are able to provide the required capabilities to fulfil future transactions.The proposed forgiveness mechanism composes the identification of forgiveness factors,information sources,and its computational model.Five positive motivations,i.e.,intent,history,apology,severity,and importance,are used to form forgiveness factors.While,forgiveness sources are information from both subjective(individual level)and objective(community level)points of view that are investigated to provide a more reliable assessment.(2)Computational model of incentive mechanism is proposed to encourage cooperation between interacting parties after a trust violation.Along with it,Zone of Forgivability is also introduced indicating some limits or boundary values to reflect the fact that the trust violation that might be forgiven should not be completely forgotten.(3)Computational model of welfare tradeoff ratio or WTR is proposed as a mechanism for restoring from trust breakdowns in distributed multi-agent systems.Welfare tradeoff ratio is a psychological variable referring to the computations in one's mind which is carried out to evaluate how much to weight the welfare of the other relative to the self.We propose computational models of metrics based on the welfare tradeoff ratio along with the way by which multiple metrics can be integrated to provide the final result.Furthermore,a number of experiments in four different environmental settings,i.e.,online community,agent-based model,iterated prisoner's dilemma,and social network,are conducted to validate the applicability of the proposed frameworks.The overall findings show that by implementing the proposed trust restoration frameworks,all simulated systems are able to improve the efficiency of agents' interactions,especially in long-term interactions and are able to deal with different levels of trust violation effectively.