主题：【海外名家讲堂】Decision rule approach to preference modelling

主讲人：波兰波兹南工业大学 Roman Słowiński教授

主持人：工商管理学院 寇纲教授

时间： 2022年1月14日（周五）14:30-16:00

举办地点：腾讯会议ID: 973342304

主办单位：工商管理学院 国际交流与合作处 科研处

主讲人简介：

Roman Słowiński is a Professor and Founding Chair of the Laboratory of Intelligent Decision Support Systems at Poznań University of Technology, and a Professor in the Systems Research Institute of the Polish Academy of Sciences. As an ordinary member of the Polish Academy of Sciences he is its Vice President, elected for the term 2019-2022. He is a member of Academia Europaea and Fellow of IEEE, IRSS, INFORMS and IFIP. In his research, he combines Operational Research and Artificial Intelligence for Decision Aiding. Recipient of the EURO Gold Medal by the European Association of Operational Research Societies (1991), and Doctor HC of Polytechnic Faculty of Mons (Belgium, 2000), University Paris Dauphine (France, 2001), and Technical University of Crete (Greece, 2008). In 2005 he received the Annual Prize of the Foundation for Polish Science - the highest scientific honor awarded in Poland, and in 2020 - the Scientific Award of the Prime Minister of Poland. Since 1999, he is the principal editor of the European Journal of Operational Research (Elsevier), a premier journal in Operational Research.

Roman Słowiński，波兰科学院波兹南分院院长，波兰最高科学奖获得者，欧洲科学院院士，波兰科学院院士，IEEE（电气和电子工程师协会）院士，国际粗糙集学会院士，欧洲运筹学会欧洲金质奖章获得者，波兹南工业大学教授、波兹南工业大学智能决策支持系统实验室的联合主席之一，现任《European Journal of Operational Research》（EJOR）主编。对决策支持技术及其方法论进行过广泛的研究，包括多准则决策辅助，偏好建模，并结合运筹学与计算智能对基于知识的决策支持进行深入研究。以作者或共同作者的身份出版14本著作，400多篇论文，其中300篇论文在主要科技杂志发表，文章Hirsch指数高达87，被引次数超过10次的文章指数高达300，总引用超过30000次。

内容简介：

The aim of scientific decision aiding is to give the decision maker(s) a recommendation concerning a set of potential actions evaluated from multiple points of view considered relevant for the decision problem at hand. These multiple points of view can be: (i) multiple voters, or (ii) multiple evaluation criteria, or (iii) multiple states of the world. They are the cornerstones of the three big sub-disciplines of decision science: (i) group decision, (ii) multiple-criteria decision making, and (iii) decision under risk and uncertainty. Their common feature is the fact that the only objective information stemming from the formulation of the corresponding decision problems is the dominance relation in the set of actions. As dominance relation is a partial weak order, it makes many actions non-comparable. To rank or classify the actions one needs to aggregate the multiple points of view, taking into account preferences of the decision maker(s) (DMs). The aggregation, which is equivalent to construction of the DMs’ preference model, is a great challenge of scientific decision aiding. In this talk, I will show how this challenge has been solved with a set of dominance-based “if..., then...” decision rules discovered from the data by inductive learning. The data are training examples showing DMs’ past decisions on a set of actions. As these examples can be inconsistent, it is convenient to structure the data prior to induction of rules using the Dominance-based Rough Set Approach (DRSA). DRSA is a methodology for reasoning about ordinal data, which extends the classical rough set approach by handling ordinal evaluations of actions and monotonic relationships between their evaluations. Since its conception, DRSA has been adapted to a large variety of decision problems. We present DRSA to preference discovery in case of multi-attribute classification, choice and ranking, in the case of evolutionary multiobjective optimization, and in the case of decision under uncertainty. The set of dominance-based decision rules has several advantages over its competitor preference models (utility functions, and outranking relations): (a) each rule is a readable scenario of a causal relationship between evaluations on a subset of attributes and a comprehensive judgment, (b) decision rules exploit ordinal information only and do not convert ordinal evaluation scales into numeric ones, (c) decision rules are non-compensatory aggregation operators, and (d) they are able to represent more complex interactions than Choquet and Sugeno integrals.

科学决策的目的是为决策者提供潜在行动的建议。决策科学的基础理论包括：（i）群体决策，（ii）多准则决策和（iii）风险和不确定性下的决策。它们的共同特点是需要从决策问题中发现评估信息的优劣关系。然而，优劣关系是部分弱序的，这使得许多评估信息不可比较。因此，为了对评估信息进行排序或分类，需要考虑决策者的偏好并对其集结。在这次报告中，我将展示如何基于决策者过去的决策偏好，通过归纳学习，从历史数据中发现决策规则。如果数据是不一致的，我们将给出基于优势的粗糙集方法（DRSA）。该方法通过处理原始评估信息和评价之间的单调关系，把经典的粗糙集方法进行了扩展，并可以应用到多属性分类、选择、排序、进化多目标优化和不确定决策等问题中。