Towards the design of robust incentive mechanisms to address subjectivity and dishonesty problems in agent reporting

Agent reporting systems, such as reputation systems and crowdsourcing platforms, provide opportunities for human users represented as agents to report their ratings or opinions via electronic channels. However, their reports could vary and are difficult to be aggregated (subjectivity problem) and they may provide untruthful reports (dishonesty problem). The two problems have attracted considerable interests from both academic and industrial societies. The first objective of this thesis is to address the subjectivity and dishonesty problems through designing three incentive mechanisms where the agents are assumed to be rational. Furthermore, considering the existence of bounded rational agents in real systems, the second objective is to evaluate the robustness of incentive mechanisms against bounded rational agents. Specifically, to address the subjectivity problem, a coalition formation game based reputation system (CONGRESS) is proposed for e-marketplaces. In CONGRESS, buyers with the same subjectivity are incentivized to form a separate \textit{club} to build seller reputation based on the ratings collected from the buyers in the club. Consequently, the proposed model avoids suffering from the bad effect of the subjectivity problem. It has been theoretically proven that buyers with the same subjectivity have the incentive to form a club if a certain condition is satisfied. Furthermore, a set of experiments has been conducted to validate the proposed reputation system. Secondly, to address the dishonesty problem, an incentive mechanism iMEMLI is proposed for e-marketplaces with limited inventory where the supply of sellers is less than the demand of buyers. In iMEMLI, more honest buyers could achieve higher utility because they have the larger probability of being allocated to the limited products. The products of more honest sellers with higher reputation are offered at higher prices. Both theoretical analysis and experimental results have shown that the proposed incentive mechanism can promote buyer honesty in providing truthful ratings and seller honesty in delivering promised products. Thirdly, to address both the subjectivity and dishonesty problems together, an incentive mechanism is proposed to elicit truthful opinions (iMET) from workers in crowdsourcing systems even when their truthful opinions are different and the majority workers behave untruthfully. iMET considers the truthful opinion difference between workers and models the workers' behaviors according to their provided opinions, and strategically rewards truthful opinions. Theoretical analysis and real-data based evaluations have shown the effectiveness of iMET in eliciting truthful opinions from workers. Finally, a simulation framework is proposed to evaluate the robustness of incentive mechanisms against bounded rational agents taking non-equilibrium strategies. A formal robustness measure is defined, inspired by the studies of evolutionary game theory. The simulation framework bases on the evolutionary dynamics to empirically quantify the proposed robustness measure, which is validated by comparing the simulation results with analytical predictions based on an extended simplex analysis approach. The simulation framework is further implemented for evaluating the robustness of several incentive mechanisms against some untruthful strategies in reputation systems for e-marketplaces, including the proposed iMEMLI incentive mechanism for EMLI. In summary, the proposed incentive mechanisms in this thesis are aimed to address the subjectivity and dishonesty problems even when the two problems coexist. The proposed simulation framework can evaluate and compare the robustness of the existing incentive mechanisms, providing insights towards the design of robust incentive mechanisms for agent-reporting systems.DOCTOR OF PHILOSOPHY (SCE