《a classication scheme for trust》由会员分享,可在线阅读,更多相关《a classication scheme for trust(10页珍藏版)》请在金锄头文库上搜索。
1、A Classification Scheme for Trust Functions in Reputation-Based Trust ManagementQing Zhang?, Ting Yu?, and Keith Irwin?Department of Computer Science North Carolina State University?qzhang4, tyu, kirwin?ncsu.eduAbstract. Reputation is an important means to establish trust in decentralized envi- ronm
2、ents such as the Semantic Web. In reputation-based trust management, an en- titys reputation is usually built on feedback from those who have direct interactions with the entity. A trust function is used to infer ones trustworthiness based on such feedback. Many trust functions have been proposed in
3、 the literature. They are typ-ically designed for specific application domains, thus differ in a variety of aspects, including trust inference methodologies, complexity and accuracy. In this paper, wepropose a classification scheme for trust functions, which will help the systematic analysis and sel
4、ection of trust functions for particular applications.1IntroductionThe Semantic Web is visioned as the next Web. By instilling semantics into ”flat” docu-ments, the Semantic Web enables users to retrieve information that satisfies queries much more complex and expressive than todays keyword-based qu
5、eries. Due to the decentralizednatureof the Semantic Web, an entity can publishwhateverin- formation without going through any scrutiny. Though seemingly relevant information can be retrieved, there is no assurance of its usefulness and correctness. Cryptographic tech-niques may be used to protect i
6、nformation confidentiality and integrity, but not its quality. To mitigate the problem,the trustworthinessof informationsourcesneeds to be considered. Reputation is an important means to establish trust in decentralized environments. An entitys reputationis usually built on feedbackfromthose whohave
7、 direct interactionswith the entity. Given a set of feedback, ones trustworthiness can be inferred through the use of trustfunctions.Trustfunctionshaveadirectimpactononesdecisionregardinginformation collection and other critical tasks, and is a key component to any reputation-based trust model.Manyt
8、rustfunctionshavebeenproposed,targetingatdifferentapplicationdomains. They differ in a variety of aspects, including trust inference methodologies, complexity and accuracy. When a user or a group of users wants to join a decentralized system, a natural question is what trust function should be used
9、for trust-related decisions. Instead ofalways designingtheirown trust functionsfromscratch, it will be beneficialif existingtrustfunctions can be reused. To make a rational choice, it is necessary to have a classification scheme for trust functions so that their advantages and disadvantages for the
10、problem at hand can be systematically analyzed.Sucha classification scheme is the focusofthis paper.In section2, we presenta general framework for reputation-based trust. The framework not only includes relevant informa- tion for trust decisions, but also models the social interaction structure of a
11、 decentralized system. The framework thus can accommodate many of the existing trust models in the lit-erature. Based on this framework, we propose a classification scheme for trust functions in section 3. The scheme has four dimensions, namely the nature of trust (subjective trust vs. objective tru
12、st), information collection (complete information vs. localized information), trust decisions (rank-based vs. threshold-based) and inputs of trust functions (opinion- based vs. transaction-based). We discuss the impact of each dimension on the properties of trust functions, and review some represent
13、ative trust functions in the literature, based onthe proposed classification scheme in section 4.This work is not the first trying to classify trust functions. For example, Ziegler etal. 15 also provides a classification scheme of trust metrics. Some of their classification dimensions overlap with o
14、urs, while some others concern about other aspects of a trust model, such as where trust computation takes place. We will compare their scheme with ours when appropriate in this paper.2A Framework for Reputation-Based TrustWe assume that entities in a decentralized environment interact with each oth
15、er through transactions. Transactions are not limited to monetary interactions. They also include ac-tivities such as retrieving information from a website, downloading files from a ftp server, etc. We further assume a transaction is uni-directional, i.e., given a transaction, there is a clear disti
16、nction between a service provider (server) and a service consumer (client). We introduce the following notion to facilitate our discussion.Trustworthiness An entitys trustworthiness is an indicator of the quality of the entitys services. It is often used to predicate the future behavior of the entity. Intuitively, if an entity is trustworthy, it is likely that the entity will provide good services in future transactions. In most trust models 4,6,11, the domain of trustworthiness is