Maintaining Preference Networks That Adapt to Changing Preferences

Abstract

Decision making can be more difficult with an enormous amount of information, not only for humans but also for automated decision making processes. Although most user preference elicitation models have been developed based on the assumption that user preferences are stable, user preferences may change in the long term and may evolve with experience, resulting in dynamic preferences. Therefore, in this paper, we describe a model called the dynamic preference network (DPN) that is maintained using an approach that does not require the entire preference graph to be rebuilt when a previously-learned preference is changed, with efficient algorithms to add new preferences and to delete existing preferences. DPNs are shown to outperform existing algorithms for insertion, especially for large numbers of attributes and for dense graphs. They do have some shortcomings in the case of deletion, but only when there is a small number of attributes or when the graph is particularly dense.