| In order to generate adaptive system responses to individual users, it is often necessary that a system establishes and maintains various hypothetical beliefs about users' task-related characteristics. This process is referred to as user modeling. An interface system equipped with user models can exhibit cooperative functionality and effective performance.;This study proposes a user modeling approach that utilizes neural networks as the knowledge base and reasoning mechanism. This approach organizes system beliefs into associative memories. This approach suggests that user modeling be a process of pattern recognition and classification, in order to capture complete and consistent profiles about users. A set of neural networks is used to associate an incomplete pattern about a user's domain knowledge with a complete hypothetical knowledge pattern that characterizes the user. Also, such patterns can be further classified into different categories in terms of the similarities.;The structure of neural network based user modeling is presented. Several different network paradigms, such as Back-propagation, ART, and Bidirection Linear Associator, are tested and integrated for pattern association and pattern classification. The experimental results are discussed in terms of the completeness, consistency, and generalization. The study shows that the proposed approach not only has the merits of conventional modeling approaches, such as fast stereotyping and assumption inheritances, but also facilitates the system performance in default reasoning, personalization, and knowledge elicitation. |
