1BC2: A True First-Order Bayesian Classifier

In previous work [3] we presented 1BC, a first-order Bayesian classifier. 1BC applies dynamic propositionalisation, in the sense that attributes representing first-order features are generated exhaustively within a given feature bias, but during learning rather than as a pre-processing step. In this paper we describe 1BC2, which learns from structured data by fitting various parametric distributions over sets and lists to the data. We evaluate the feasibility of the approach by various experiments.