Abstract

The vast amount of information on the Internet drives the Information Age: we want to know more about more. Unfortunately, the Internet does not guarantee the quality of this information. It is posted by newspapers, scientific institutes, salesmen and individuals. All of these people have their own interests, their own cultural and political background and their own objective for posting this information. Therefore, when gathering information from the Internet we have to be careful about which of these sources to use: we only want the trusted ones. And because we want to gather as much information as possible, an automated way of 'harvesting' it was called for. This application should be able to gather trusted information from all over the Web and label its quality in the sense of trustworthiness.
This thesis presents the blueprint for such an application including a prototype the can indicate trust levels for commercial websites. The prototype is based on the way people assess trust on websites. This was investigated and formalised into a model. Every user of this prototype has his own agent that has learned what its user finds important in trust. The agent examines the site and returns the trust value the user would have given. When this user wants to have the trust opinion of a friend or an expert he can use his agent too. All the agents will negotiate with each other on the outcomes and supply the user with one uniform result. In this thesis tests have been performed on the quality of these agents and the way in which they should be trained.