Towards Scalable Scoring for Preference-based Item Recommendation

Preference-based item recommendation is an important technique employed by online product catalogs for recommending items to buyers. Whereas the basic mathematical mechanisms used for computing value functions from stated preferences are relatively simple, developers of online catalogs need flexible formalisms that support the description of a wide range of value functions and map to scalable implementations for performing the required filtering and evaluation operations. This paper introduces an XML language for describing simple value functions that allow emulating the behavior of commercial preference-based item recommendation applications. We also discuss how the required scoring operations can be implemented on top of a commercial RDBMS, and present directions for future research.