Vorschau

PDF, Englisch Download (5MB) | Nutzungsbedingungen

Abstract

Services are ubiquitous. In daily life, we can find service provisioning everywhere such as online shopping, online storage, doctor, hotel, lawyer, restaurant, etc. With the development of Web 2.0 technology, a huge amount of information about services has become available on the Internet. For instance, on a review website people can discuss which restaurant serves the best Chinese food; in a blog, an author posts an article about the experience of visiting a doctor. The abundance of services and the overload of service information online, result in two main problems. The first problem is service selection; the second one is the overload of consumer-driven information which refers to information such as reviews, articles, assessments, and discussions generated by service consumers.

The concept of trust is proposed to solve the two problems. The computational concept of trust is defined as a subjective probability, which makes a prediction of the occurrence of an event such as a good service provisioning. Software used for building and managing trust data related to service offerings, is called Trust Management System (TMS). The first topic is trust model. A trust model is the computing kernel of a TMS that calculates the trust value of a service. Another significant topic regarding trust management for services is the robustness of a TMS. Robustness of a TMS refers to the ability of a TMS to cope with inaccuracy (deliberate or accidental) in the consumer-provided information used for computing trust. There are many trust models that have been proposed. I do not know of any survey analyzing and comparing different trust models with respect to trust in services. In this thesis, 40 trust models are compared from both a theoretical and a practical perspective, using criteria such as application context, information representation, properties of trust evaluation, and robustness of system. In addition, a trust model framework for service provisioning is proposed. This framework is considered a meta-model covering all existing trust models. A concrete trust model can be derived by instantiating the meta-model.

In the thesis, four concrete services which cover both quantitative and qualitative services are studied. A quantitative service refers to a service the quality of which can be measured objectively. For a qualitative service there is no general agreed-upon objective measure for service quality. The first case study is about Online File Storage Service (OFSS) which is categorized as a quantitative service. The trust model, R-Rep, for a OFSS is proposed. In order to mitigate manipulation, a statistics based detection mechanism, named Baseline Sampling (BS), is introduced. In addition, when social network information among users is available, Clique Identification (CI) is used to detect manipulative groups. One e-commerce website, Taobao.com, and two review websites, TripAdvisor.com and Dianping.com, are chosen as case studies for trust building and managing in the context of qualitative service. For each case, specific trust models which consider intrinsic robustness enhancement by designing special weight functions are proposed. Meanwhile, machine learning-based extrinsic robustness enhancement is applied. Three types of machine learning approaches, clustering, classification and Annotation-Auxiliary Clustering (AAClust), are applied to identify manipulative behavior.