Please use this identifier to cite or link to this item:
|Title:||Trust beyond reputation: novel trust mechanisms for distributed environments||Authors:||Liu, Xin.||Keywords:||DRNTU::Engineering::Computer science and engineering::Computer systems organization::Computer-communication networks||Issue Date:||2011||Source:||Liu, X. (2011). Trust beyond reputation: novel trust mechanisms for distributed environments. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||In recent years, people have taken a more active role in participating in massively distributed systems (e.g., Web 2.0 applications) by intensively interacting with other system participants. Such rich interactions enhance Internet experience significantly. However, due to the characteristics of such systems like openness, scale, heterogeneity, etc., it is difficult to ensure smooth/reliable interactions among the participants. Trust management, which aids automated decision-making process or alternatively provides decision support, is a popular approach to help participants to select the trustworthy interaction partners. Traditional reputation based trust mechanisms may be effective in certain scenarios, but they suffer from two main issues: (i) they often rely upon knowledge (i.e., historical information about past behavior of a specific agent) that may not actually be available or accurate to the assessor. (ii) Even if the required historical information is available, current mainstream methods simply investigate outcomes of the target agent's past transactions to derive a single trust/reputation value (and sometimes, an additional confidence metric on that value), thus failing to detect dynamic behavior of the intelligent agents who may strategically change his behavior to maximize profits. It is thus essential to develop novel mechanisms that support ``trust beyond reputation''. For the issue (i), we first introduce a stereotyping based computational trust model StereoTrust. The trustor (i.e., the agent that needs to derive trust on other agents) forms stereotypes using his previous transactions with other relevant agents. When facing a stranger, the stereotypes matching stranger's profile are aggregated to derive the expected trust. StereoTrust uses an intuitive and heuristic method to derive and combine stereotypes for trust assessment. In order to improve StereoTrust by discriminating different types of stereotypes, we then develop MetaTrust, a machine learning techniques based trust framework. The agent uses his own previous transactions (with other agents) to build a knowledge base, and utilize this to assess the trustworthiness of a transaction based on the associated features that are harnessed using machine learning techniques. In addition, we propose to construct a local knowledge sharing overlay network (LKSON) to bootstrap the inexperienced agents who have limited local knowledge. In order to validate usefulness of the stereotype based trust mechanisms in reality, we apply them to two different massively distributed applications: (1) Peer-to-Peer backup storage systems (i.e., a decentralized setting) for selecting suitable data holders and (2) Internet online auction site (i.e., a centralized setting) for detecting auction frauds. Simulation results show that in both settings, our approaches work effectively and efficiently in comparison to other existing approaches. For the issue (ii), we develop solutions to capture dynamic behavior of the target agent based on his past transactions in two aspects: (1) we propose a context matching based approach to detect the target agent's past behavior patterns for trust assessment; (2) we propose the concept of imprudence to study and detect the inappropriate behavior of an agent with high reputation value and relatively long history, and hence considered `reliable' by traditional reputation systems.||URI:||http://hdl.handle.net/10356/46445||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.