The Representation, Acquisition And Inference Of Personalized Requirements: A Case Study

With the continuous growth and proliferation of the web based information systems, personalization has emerged as a critical application which is essential to the success of a web site. Personalization can be defined as any action that tailors the web experience to a particular user, or set of users. The principle elements of the personalization system include modeling the web objects (resources to be personalized such as web pages, research papers) and subjects (users), matching between and across objects and subjects.This dissertation aims at the key techniques in the personalization, especially the techniques of user interests (preferences) representation, acquisition and use of the learned model in the recommendation process. We use domain ontology and XML to represent resources and user profiles, and study the learning algorithms used to build the user profile with the user usage data. We suppose that these approaches offer advantages over traditional keywords based profile representation in the context of personalization systems. The main points of our work are described as follows:Ontology-XML based personalization system architecture.In order to personalize different types of complex objects with underlying properties and attributes, we introduce the domain ontology to the personalization system and propose an Ontology-XML based personalization system. The use of ontology can lead to deeper interaction between user and web site and can provide the capability to explain and reason about user actions. We give the ontology representation framework and objects representation method using XML.Ontology-XML based user interest modelRepresenting and modeling user interests is the key to the personalization systems. We define the user interest as a pseudo instance of the resource class which means that each concept or feature in the user model has a weight value. In order to learn the user interest model based on the user's feedback, we represent the entire rating mechanism as feed-forward neural network model and adopt the generalized delta rule as basic learning schema by modifying it to our framework. Using semantic knowledge in user modeling will make the user model more precise and more understandable.Mining the user interest rules using reduction ideaWe propose a new method using semi-structure rules to represent user interests based on...