Uncertainty Modelling for Adaptive Information Management

The management of complex systems strongly depends on the ability to handle huge amounts of information. The experience accumulated on a problem represents knowledge we would like to capitalise on in the future and information retrieval (IR) systems offer a valuable support. Uncertainty is a fundamental component of the description of a piece of data and its explicit modelling is the purpose of our work. In a standard IR context, uncertainty permeates the behaviour of both the system and the users and we investigate the effects of its explicit modelling on classical IR parameters like precision and recall. We present a keyword based model that, capitalising on the flexibility of fuzzy sets, extends the traditional two dimensional vector approach to data abstraction evolving it into a paradigm where relevance is tightly coupled with uncertainty and the view the system has on data evolves dynamically through an adaptivity process. A prototype system (DUNE) has been derived from the general model and we investigate its applications on knowledge management aspects of a help desk system.