CAREER: Mining Reliable Information from Crowdsourced Data

随着移动设备和社交媒体平台的激增，任何人都可以随时随地发布对任何活动、事件或对象的观察。这些巨大的众包数据的汇合可以促成一个廉价、可持续和大规模的决策系统，这是以前从未可能实现的。这样的系统可以极大地提高交通、医疗保健和许多其他应用的效率和成本。建立这样一个系统的主要障碍在于信息的准确性问题，即个人用户可能提供不可靠甚至误导性的信息。该项目确定了在从嘈杂和不可靠的众包数据中挖掘可靠信息的任务中的重要研究问题，并进行了一项综合研究和教育计划来解决这些问题。该项目通过整合不同来源的数据，解决了信息的准确性问题，这将有利于许多应用程序，其中众包数据无处不在，但准确性可能受到质疑。特别是，该项目通过考虑众包的各种属性，开发了挖掘可靠信息的新方法：1)众包平台收集用户对特定对象的观察。其他有价值的信息源，如时空、用户影响和文本数据，被用来有效地从这些观测中检测可靠的信息。2)设计有效的隐私保护和预算分配机制，以更好地激励积极的众包。这些研究与对所提出方法的理论和实践方面的探索是相结合的。从理论上探讨了估计可靠性的置信度和所提方法的收敛等基本问题。从实践的角度来看，建议的方法适用于解决交通、医疗保健和教育等各种应用中的挑战性问题，以使人们能够对这些领域有新的见解。除了研究进展之外，该项目还有助于教育创新，因为所建议的方法被应用于同行评估和问题回答等教育方法。有关该项目的其他信息，包括研究结果、出版物、数据集和软件，请访问http://www.cse.buffalo.edu/~jing/crowd.htm

With the proliferation of mobile devices and social media platforms, any person can publicize observations about any activity, event or object anywhere and at any time. The confluence of these enormous crowdsourced data can contribute to an inexpensive, sustainable and large-scale decision system that has never been possible before. Such a system could vastly improve the efficiency and cost of transportation, healthcare, and many other applications. The main obstacle in building such a system lies in the problem of information veracity, i.e., individual users might provide unreliable or even misleading information. This project identifies important research questions in the task of mining reliable information from noisy and unreliable crowdsourced data, and pursues an integrated research and education plan to address these questions. Through integrating data from various sources, this project addresses information veracity, which will benefit the many applications where crowdsourced data are ubiquitous but veracity can be suspect.In particular, this project develops novel methods to mine reliable information by taking into consideration various properties of crowdsourcing: 1) Crowdsourcing platforms collect users' observations about certain objects. Other valuable information sources, such as spatial-temporal, user influence, and textual data, are leveraged to effectively detect reliable information from these observations. 2) Effective privacy protection and budget allocation mechanisms are designed to better motivate active crowdsourcing. These investigations are integrated with the exploration of both theoretical and practical aspects of the proposed methods. From the theoretical perspective, fundamental questions regarding the confidence in the estimated reliability and the convergence of the proposed methods are explored. From the practical perspective, the proposed methods are adapted to tackle challenging problems in various applications such as transportation, healthcare and education to enable new insights into these domains. In addition to the research advances, this project contributes to educational innovation, as the proposed methods are applied to educational methodologies such as peer assessment and question answering. Additional information about this project, including research results, publications, datasets, and software, can be found at http://www.cse.buffalo.edu/~jing/crowd.htm