SBIR Phase I: Relational Bayesian Modeling for Electronic Commerce

This Small Business Innovation Research Phase I Project will advance Relational Bayesian Modeling (RBM) technology, with specific focus on Information-based Technology applications in electronic commerce, especially customer and application-system behavior modeling. Relational Bayesian modeling combines recent advances in graphical statistical modeling with modern relational and object-oriented data models. An RBM is compact, precise, and efficient. It is compact because structural information, represented graphically, identifies a minimal set of numeric information needed to complete the model. It is precise because any probabilistic model, of arbitrary complexity, can be represented. No global simplifying assumptions are made. Finally, the structural information can be exploited to support efficient computation with the model. Work under this SBIR will focus an open problem in RBMs: the representation and discovery of, and computation with, probabilistic dependencies in many-to-one relations. Preliminary research has shown that relational Bayesian models can effectively capture dynamic behavior, and can be applied to recognize behavioral profiles in real-time, of both users and systems. This will enable a new generation of adaptive web interfaces, improve the reliability and predictability of multi-tier system performance, and be an important element in realizing autonomic computing. The commercial application for this technology is electronic commerce which is in need of the improve reliability and predictability of multi-tier systems which this solution will provide.