Eliciting and Using Goal Variability

目标模型是对需求工程中涉众意图进行建模和推理的一种流行方法。在早期需求分析期间，当为未来的系统寻找正确的特性集时，以及在运行时，当涉众的需求必须转换为运行系统的配置时，它们被特别建议用于决策制定。在本建议中，我们将重点放在三个方向上的目标模型的研究上：(a)它们的可理解性和作为可视化工件的使用，(b)它们从利益相关者那里得到的启发，以及(c)它们在软件配置和定制中的利用。关于可理解性，我们建议通过实验研究目标模型作为视觉工件的价值，以促进对最优解的临时识别。我们对贡献链接特别感兴趣，目标模型中的元素表示一个涉众目标的满足如何影响另一个涉众目标的满足。我们想研究什么样的方法来表示贡献更适合于对最优解的临时探索。我们还建议将目标模型与其他决策支持可视化进行比较，并研究我们过去提出的与自动推理工具的交互是否有助于理解。关于启发，我们感兴趣的是目标模型的两个基本组成部分是如何产生的：前面提到的贡献元素和诱发变量的分解。在贡献因素方面，我们首先提出研究假设情境和认知偏差在其激发过程中的作用。我们的早期实验表明，这些因素可以影响引出贡献度量的有效性。我们还建议研究在获得可靠的贡献度量方面，选择启发工具（例如各种版本的定性与定量量表）的作用。此外，诱导变异的分解是引入变异（即决策）点的目标的细化。在过去，我们已经介绍了一种系统的方法来获得这样的变异点，我们现在希望根据它所导致的变异的质量和完整性来经验地评估它。在软件配置和定制方面，我们建议继续我们目前在高度可定制系统的设计以及现有系统的配置方面的工作。关于设计高度可定制的系统，我们建议扩展我们早期在基于策略的行为定制方面的工作，并定义更严格的方法来开发根据直接来自目标和其他需求模型的策略对象进行行为的系统。针对现有系统，我们建议研究概念建模和自动推理技术的应用，以促进大型复杂信息系统的配置。我们还建议研究常见的流行系统和使用数据挖掘技术来识别大量用户中的配置模式，旨在使用这些模式来通知和促进配置过程。建议的研究针对目标建模的各个方面，我们发现这些方面对于在行业中采用目标建模很重要。我们相信，关于我们如何引出有效的目标模型并将其有效地用于问题探索和理解的具体证据将使它们对实践者更具吸引力。同样，为定义和控制软件系统中的可变性引入基于需求的工具集，既使定制过程更易于访问和有效，又允许软件工程师显著地增加其系统的适应性。

Goal models are a popular approach for modeling and reasoning about stakeholder intentions in requirements engineering. They have particularly been proposed for use in decision making both during early requirements analysis, when the right set of features for a future system is sought, and at run-time, when stakeholder needs must translate into configurations of a running system. In this proposal we focus on the study of goal models along three directions: (a) their comprehensibility and use as visual artifacts, (b) their elicitation from stakeholders and (c) their utilization in software configuration and customization. With respect to comprehensibility we propose to experimentally study the value of goal models as visual artifacts that facilitate ad-hoc identification of optimal solutions. We are particularly interested in contribution links, elements in the goal models that represent how satisfaction of one stakeholder goal influences satisfaction of the other. We want to study what ways to represent contributions is more suitable for ad-hoc exploration of optimal solutions. We also propose to compare goal models with other decision support visualizations as well as to study whether interaction with automated reasoning tools we have proposed in the past facilitates comprehension. With respect to elicitation, we are interested in how two basic components of goal models come about: the aforementioned contribution elements and variability-inducing decompositions. In terms of contribution elements, we firstly propose to study the role of assumed context and cognitive biases during their elicitation. Our early experiments suggest that such factors can influence the validity of the elicited contribution measures. We also propose to study the role of the choice of elicitation instrument (e.g. various versions of qualitative vs. quantitative scales) in acquiring reliable contribution measures. Variability-inducing decompositions, moreover, are refinements of goals that introduce variation (i.e. decision) points. In the past we have introduced a systematic approach to acquiring such variation points, which we now wish to empirically evaluate in terms of the quality and completeness of the variability it results to. In terms of software configuration and customization, we propose to continue our current work on the design of highly-customizable systems as well as the configuration of existing systems. With respect to designing highly-customizable systems we propose to extend our earlier work on policy-based behavioral customization and define more rigorous ways for developing systems that behave according to policy objects that are directly derived from goal and other requirements models. In terms of existing systems, we propose to study application of conceptual modeling and automated reasoning techniques in order to facilitate configuration of large and complex information systems. We also propose to study common popular systems and the use of data mining techniques for identifying configuration patterns within large numbers of users, aiming at using those patterns for informing and facilitating the configuration process. The proposed research is targeted towards aspects of goal modeling that we find important for enabling its adoption in the industry. We believe that concrete evidence on how we can elicit valid goal models and use them effectively for problem exploration and understanding will make them more appealing to practitioners. Likewise, introducing requirements-based toolsets for defining and controlling variability in software systems both makes the customization process more accessible and efficient and allows software engineers to dramatically increase the adaptability of their systems.