Learning preferences in Qualitative Choice Logic and some of its variants: an application for antibiotics recommendations

Qualitative Choice Logic (<i>QCL</i>) is a non-classical logic for representing and reasoning with preferences. It adds to classical propositional logic a new connective called ordered disjunction (×→). x×→y intuitively means: if possible <i>x</i>, but if <i>x</i> is not possible then at least <i>y</i>. Different variants of <i>QCL</i> have been proposed. Among them, we cite Prioritised Qualitative Choice Logic, Conjunctive Choice Logic and Lexicographic Choice Logic. In this paper, we propose a method for learning preferences in the context of <i>QCL</i>. The method is based on an adaptation of association rules based on APRIORI algorithm. The adaptation consists of (1) generating rules that have propositional formulas in their antecedent instead of itemsets and (2) using variations of the support and confidence measures according to the semantics of <i>QCL</i>. We show that <i>QCL</i> is fully adapted for modelling experts reasoning for providing recommendations of antibiotics. Another contribution of the paper concerns a generalisation of the proposed method for learning preferences in the context of <i>QCL</i> variants.

定性选择逻辑（Qualitative Choice Logic，QCL）是一类用于表示偏好并开展偏好推理的非经典逻辑。它在经典命题逻辑的基础上新增了一种名为有序析取（×→）的连接词。x×→y的直观语义为：若条件允许则优先选择x，若x不可行则至少选择y。目前已有多种QCL变体被提出，其中包括优先定性选择逻辑、合取选择逻辑与字典序选择逻辑。本文提出了一种在QCL语境下学习偏好的方法，该方法基于对基于APRIORI算法的关联规则的适配改造，具体改造内容分为两点：其一，生成前件为命题公式而非项集的关联规则；其二，结合QCL的语义调整支持度与置信度两类度量指标的计算方式。研究证明，QCL完全适配于建模专家推理流程以生成抗生素用药推荐。本文的另一项贡献在于，将所提出的偏好学习方法推广至各类QCL变体的应用场景中。