Approximate Contextual Preference Scoring in Digital Libraries

The increasing availability of a wide variety of digital libraries provides a heterogeneous population of users with access to a heterogeneous collection of data. To help users find interesting items from this huge volume of available information, personalization systems allow users to directly or indirectly indicate preferences. Such preferences may depend on context, for instance, on the device used to access information. In this paper, we consider a preference model that allows users to assign interest scores to pieces of data based on context. We address the problem of locating interesting data items efficiently by exploiting some form of precomputation. To this end, instead of computing scores for all data items in all potential context states, we exploit the hierarchical nature of context attributes to identify representative context states and compute scores for them. This provides users with fast but potentially approximate results. We discuss this issue and report some preliminary results.