Observing web users: conjecturing and refutation on partial evidence

Abstract

Personalized hypermedia and Web systems are confronted with the challenge of inferring complex user traits like knowledge or preferences from very basic data like the 'clickstream' or ordinal-scale ratings. In consequence, the resulting user models are only approximations that must be subject to continuous revision. Nonetheless, knowledge revision procedures are rarely made explicit in existing adaptive systems and models. In this paper, we sketch a frame-work for user modeling structured around revision and refutation of provisional conjectures drawn from basic data. This model can be used as a reference framework for the evaluation of the adequacy of the inferences carried out by existing adaptive hypermedia systems. Additionally, a number of existing adaptive systems is reviewed according to the core concepts of this model. It is also argued that Possibility Theory can be used to generalize different forms of uncertainty that are not precisely justified in existing applications.