Developing Preference Band Model to Manage Collective Preferences

Discovering user preference is an important task in various database applications, such as searching product information and rating goods and services. Previous work addressing user preference in databases either assumes that users are able to formulate explicit preference criteria when querying, which may not be feasible in reality, or develops mining techniques to discover users’ implicit preference from their track record of purchasing and then to generate a restrictive linear ranking on items, which needs to process a huge amount of data possibly with noise. However, there still lacks of a unifying model that is able to capture both implicit and explicit user preference information and to support managing, querying and analysing the information obtained from different sources. In this paper, we present a framework based on our newly proposed Preference Band Model (PBM), which aims to achieve several goals. First, the PBM can serve as a formal basis to unify both implicit and explicit user preferences. We develop the model using a matrix-theoretic approach. Second, the model provides means to manipulate different sources of preference information. We establish a set of algebraic operators on Preference-Order Matrices (POMs). Third, the model supports direct querying of collective user preference and the discovery of a preference band. Roughly, a preference band is a ranking on sets of equally preferred items discovered from a POM to classify collective user preference. We demonstrate the applicability of our framework by studying two real datasets.