Intelligent Group Decision Systems and Preference Learning

最近的技术进步（例如，互联网、电子邮件和在线社交网络）将信息和人联系在一起，但也加速了选择和信息的爆炸性增长。这种信息和选择的过载使决策变得更加困难，这可能会使从简单的日常生活决策到高风险商业决策的决策变得瘫痪。为了克服这种决策复杂性，智能决策系统（例如，推荐系统）将在不久的将来发挥至关重要的作用，以帮助或自动化我们的决策过程。任何智能决策系统的主要工作之一就是了解最终做出决策的用户偏好。当前的系统（例如，推荐系统）要求用户不断地报告他们的偏好（例如，评价电影、书籍等）。 这一要求给用户带来了很高的认知负担，从而降低了他们的效率。 该研究计划的偏好学习方法是通过要求用户提供尽可能少的明确信息来最大限度地减少这种负担，以使智能决策系统实用并帮助保护个人隐私。由于通过在线社交网络、电子市场或任何其他Web或移动的应用程序生成的用户行为数据的可用性不断增加，这种方法是可能的。 这个建议的重点是群体决策问题，这涉及到为一组有自己的个人和可能相互冲突的偏好的个人做出决策。群体决策问题是普遍存在的（例如，社会团体、商业组织、公共政策等的决策）。群体决策中的偏好学习问题更为突出，因为决策者需要学习群体偏好的集合，这可能是具有挑战性的。该研究计划的长期目标是开发实用的群体决策算法，并对其进行经验和理论评估。该计划侧重于偏好学习-通过减少决策所需的明确信息-和强大的计算效率算法的设计，可以处理不确定和嘈杂的偏好。将研究各种群体决策问题，包括投票，匹配和分配。该计划包括几个关键的方法论主题，如不完全偏好的随机优化，偏好学习的概率模型，以及群体决策算法性能的理论和实证分析。该计划的产出将有助于为智能决策系统铺平道路，以协助个人，组织，企业和国家在医疗保健，娱乐，公共政策，营销和金融等众多领域的群体决策。

Recent technological advances (e.g., Internet, email, and online social networks) have united information and people, but have accelerated the explosive growth of choices and information. This information and choice overload has made decision making harder, which can paralyze decision making for tasks ranging from simple daily-life decisions to high-risk business decisions. To overcome this decision-making complexity, intelligent decision systems (e.g., recommender systems) will play a crucial role in the near future to assist or automate our decision-making processes.One of the main endeavors in any intelligent decision system is learning the user preferences upon which decisions are ultimately made. Current systems (e.g., recommender systems) require users to constantly report their preferences (e.g., rate movies, books, etc.). This requirement imposes high cognitive burden on users, and consequently reduces their effectiveness. This research program's approach to preference learning is to minimize this burden, by requiring as little explicit information as possible from users to make intelligent decision systems practical and help preserve individual's privacy. This approach is possible due to the increasing availability of user behaviour data generated through online social networks, e-marketplaces, or any other Web or mobile applications. The focus of this proposal is group decision problems, which involve making decisions for a group of individuals who have their own personal and possibly conflicting preferences. Group decision problems are prevalent (e.g., decisions for social groups, business organizations, public policies, etc.). The problem of preference learning in group decision making is more predominant as the decision maker needs to learn a collection of group preferences, which can be challenging to obtain. The long-term objective of this research program is the development of practical group decision making algorithms, and their empirical and theoretical assessment. The program focuses on both preference learning---by reducing the explicit information needed for decision making---and the design of robust computationally-efficient algorithms, which can handle uncertain and noisy preferences. A variety of group decision problems will be studied, including voting, matching, and assignment. The program encompasses several key methodological themes such as stochastic optimization for incomplete preferences, probabilistic models for preference learning, and theoretical and empirical analyses of performance of algorithms for group decision making. The output of this program will help pave the way for intelligent decision systems to assist group decision making for individuals, organizations, businesses, and nations in a myriad of domains including health care, entertainment, public policy, marketing, and finance.