Modelling Probabilistic Inference Networks and Classification in Probabilistic Datalog

摘要

Probabilistic Graphical Models (PGM) are a well-established approach for modelling uncertain knowledge and reasoning. Since we focus on inference, this paper explores Probabilistic Inference Networks (PIN's) which are a special case of PGM. PIN's, commonly referred as Bayesian Networks, are used in Information Retrieval to model tasks such as classification and ad-hoc retrieval. Intuitively, a probabilistic logical framework such as Probabilistic Datalog (PDatalog) should provide the expressiveness required to model PIN's. However, this modelling turned out to be more challenging than expected, requiring to extend the expressiveness of PDatalog. Also, for IR and when modelling more general tasks, it turned out that 1st generation PDatalog has expressiveness and scalability bottlenecks. Therefore, this paper makes a case for 2nd generation PDatalog which supports the modelling of PIN's. In addition, the paper reports the implementation of a particular PIN application: Bayesian Classifiers to investigate and demonstrate the feasibility of the proposed approach.