Building robust trust model for multi-agent systems

• Main Author: Weng, Jian Shu
• Other Authors: Miao, Chun Yan
• Format: Theses and Dissertations
• Language: English
• Published: 2008
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/13588
• Institution: Nanyang Technological University

In multi-agent systems, modeling of trust equips agents with the ability to establish trust in one another based on their personal previous interaction experiences. In the case that agents’ personal interaction experiences are not sufficient to establish trust, third-party testimonies are usually sought and aggregated. However, the presence of unfair testimonies deteriorates the performance of trust models. Unfair testimonies are testimonies deviating from agents’ real behaviors that are perceived by the testimony receivers. The mitigation of the adverse effects of unfair testimonies is a fundamental problem in the research on trust modeling in multi-agent systems. This thesis presents the research into making trust models robust in the presence of unfair testimonies. It proposes an uncertainty-based testimony-filtering method to mitigate the adverse effects of unfair testimonies. The proposed filtering method is then revamped with a novel credibility model. The revamped model not only improves the effectiveness of the filtering method, but is also general enough to be applied to the existing trust models. Unlike most of the existing methods, the proposed credibility model does not require any additional mechanism or knowledge other than the testimonies shared among the agents. Empirical evaluations also show that our proposal consistently outperforms related work. This work also goes beyond the existing methods in that it discards the common assumption of the existing methods that each agent would obtain testimonies from all other agents directly in the system. Instead, we propose a credibility-aware referral process on top of the credibility model. The credibility-aware referral process facilitates agents’ testimony discovery in an efficient manner, in which more credible agents are iteratively requested to discover testimonies by testimony discovery initiator. Furthermore, this thesis proposes an approach to counteract malicious referrers during testimony discovery. The presence of malicious referrers aggravates the adverse effects of unfair testimonies. This effect is simply ignored in the existing methods. The proposed research can be applied in many application domains. Some typical domains include (1) e-commerce systems such as the online auction website ebay.com where people can transact with other people geographically located thousands of miles away, (2) Peer-to-Peer content sharing networks with which digital contents can easily be shared among users widely distributed in different locations around the world, (3) the Grid with which various computing resources can be shared among various users and institutions, and (4) massive multi-player online role playing games (MMORPGs) in which various user avatars interact with each other.