In this thesis, we address the issue of uncertainty in Object-Oriented Business Rules Management Systems (OO-BRMSs). To achieve this aim, we rely on Probabilistic Relational Models (PRMs). These are an object-oriented extension of Bayesian Networks that can be exploited to efficiently model probability distributions in OO-BRMSs.
It turns out that queries in OO-BRMS are numerous and we need to request the PRM very frequently. The PRM should then provide a rapid answer. For this reason, we propose, in the first part of this thesis, a new algorithm that respects two specificities of OO-BRMSs and optimizes the probabilistic inference accordingly. First, OO-BRMSs queries affect only a subset of their base, hence, the probabilities of interest concern only a subset of the PRM random variables. Second, successive requests differ only slightly from each other. We prove theoretically the correctness of the proposed algorithm and we highlight its efficiency through experimental tests.
During the second part, we establish general principles for probabilistic OO-BRMSs and we describe an approach to couple them with PRMs. Then, we apply the approach to IBM Operational Decision Manager (ODM), one of the state-of-the-art OO-BRMSs, and we provide an overview of the resulted prototype.
Finally, we discuss advanced techniques to compile elements of ODM technical language into instructions that are exploitable by the PRM probabilistic engine and we provide some directions for future works.