EAGER: Decision Support System for Reasoning with Preferences

自动化决策支持系统帮助用户在一组备选方案中做出明智的选择，同时考虑到用户偏好和多个系统属性之间的权衡。在软件工程领域，决策支持系统被用来根据数量偏好和权衡来帮助评估可选择的设计、技术和管理选择。通过直接向利益相关者询问每个属性的感知效用/价值，或者根据过去的经验和专业知识对这种效用/价值进行量化，来评估对替代方案的偏好。然而，在大多数实际设置中，对属性的偏好不能全部量化。另一方面，仅定性地考虑偏好（将它们指定为选项之间的简单相对顺序）也是不实际的。为了克服这些限制，建议的研究侧重于开发决策支持系统的新范式，其中以定性和定量的方式指定偏好。这项工作的主旨将是：(a)开发强大的形式主义，以统一的方式用定量和定性偏好表示和推理，(b)调查应用领域特定的形式主义扩展，以及(c)确定决策支持系统作为偏好分析器的实际应用的实施策略。预期的结果将有助于在多个领域实现特定于应用程序的健壮的决策支持系统，包括产品线工程、安全关键系统开发和面向目标的需求工程，通过支持关于偏好的改进的自动推理。这项工作将有助于在软件工程、形式化方法和人工智能技术方面培养一名博士后学者和一名研究生。研究成果将通过期刊和会议的出版物传播。

Automated decision support systems help users make informed and intelligent choices over a set of alternatives, taking into account user preferences and trade-offs among multiple system attributes. In the software engineering domain decision support systems are used to help in evaluating alternative design, technical and managerial choices in terms of quantitative preferences and trade-offs. Preferences over alternatives are evaluated either by directly soliciting from the stakeholders a measure of the perceived utility/value of each attribute, or by quantifying such utility/value based on past experience and expertise. In most practical settings, however, preferences over attributes cannot all be quantified. On the other hand, considering preferences only qualitatively (specifying them as simple relative orderings between alternatives) is also not practical. To overcome these limitations, the proposed research focuses on developing a new paradigm for decision support systems, where preferences are specified both in qualitative and quantitative terms.The main thrust of this work will be to: (a) develop robust formalisms for representing and reasoning with quantitative and qualitative preferences in an unified fashion, (b) investigate application-domain specific extensions to the formalisms, and (c) identify implementation strategies for practical application of the decision support system as a preference analyzer. The anticipated results will help realize application-specific robust decision support systems in multiple domains, including product-line engineering, safety-critical system development, and goal-oriented requirements engineering, by enabling improved automated reasoning about preferences. This work will contribute to research-based training of a postdoctoral scholar and a graduate student in techniques that cut across software engineering, formal methods and artificial intelligence. Research results will be disseminated through publications in journals and conferences.