Computational Preference Assessment and Optimization for Group Decision Support

信息技术的进步所提供的丰富的数据和选择范围深刻地改变了个人、专业人士和政策制定者所面临的决策的性质。智能决策支持系统（DSSs）对于帮助管理这种复杂性至关重要。不幸的是，DSSs需要相当多的（个人或组织）偏好知识，导致“偏好瓶颈”：我们如何以自然、不引人注目的方式引出偏好？在群体决策（或社会选择）问题中，偏好瓶颈被加剧，因为决策支持系统必须汇总个体群体成员的偏好。这不仅增加了所需偏好信息的数量，而且还为个人提供了歪曲其偏好的动机，从而损害了社会目标。本研究项目将为社会选择（群体）环境下的决策支持开发新的计算模型和算法，重点研究偏好的有效评估和激发，以帮助打破偏好瓶颈。具体目标包括开发：(a)具有不完全偏好的决策算法，以及有效的、最少的偏好引出；(b)在现实的信息假设下对误报偏好的动机进行理论和实证分析的新技术；(c)用于学习用户偏好统计模型的新算法，可以通过我们的优化和启发方法加以利用。这些技术将应用于一系列社会选择问题，包括投票、匹配和分割问题。一个研究重点将涉及发现和利用社会网络中的偏好相关性的方法，以改善部分偏好的启发和群体决策。本研究项目的结果将对大多数形式的群体决策/社会选择产生广泛的影响，包括信息高效的投票方案、市场细分工具、匹配机制、政策设计和群体推荐系统。

The abundance of data and range of options afforded by advances in information technology have profoundly changed the nature of the decisions that face individuals, professionals and policy makers. Intelligent decision support systems (DSSs) are vital to help manage this complexity. Unfortunately, DSSs require considerable knowledge of (individual or organizational) preferences, leading to the "preference bottleneck": how do we elicit preferences in a natural, unobtrusive manner? In group decision making (or social choice) problems, the preference bottleneck is exacerbated, since a DSS must aggregate the preferences of individual group members. This not only increases the volume of preference information needed, but it provides incentives for individuals to misrepresent their preferences to the detriment of the social objective.This research program will develop new computational models and algorithms for decision support in social choice (group) settings, with a focus on the effective assessment and elicitation of preferences to help break the preference bottleneck. Specific objectives include development of: (a) algorithms for decision making with incomplete preferences, and effective, minimal elicitation of preferences; (b) new techniques for the theoretical and empirical analysis of the incentives for misreporting of preferences under realistic informational assumptions; and (c) new algorithms for learning statistical models of user preferences that can be exploited by our optimization and elicitation methods. These techniques will be applied to a range of social choice problems, including voting, matching, and segmentation problems. One research thrust will involve methods to discover and exploit preference correlations in social networks to improve elicitation and group decision making with partial preferences. The results of this research program will have broad consequences for most forms of group decision making/social choice, including informationally efficient voting schemes, market segmentation tools, matching mechanisms, policy design, and group recommender systems.