CAREER: Human-Centric Knowledge Discovery and Decision Optimization

人类既是大数据的生产者，也是大数据的消费者。因此，对人类在挖掘大数据过程中的作用进行建模，对于释放挖掘知识在健康、教育、安全和科学发现等各个重要领域的巨大潜力至关重要。本研究的目的是建立一个以人为中心的学习框架，利用人类产生的大数据和计算模型的力量。这个研究项目的重点还在于将进行的研究活动纳入教材，以便在资料检索和数据挖掘领域培训和教育学生。此外，提供数据分析方法和应用的专门培训，以加强对非stem和社区大学生的大数据相关技术教育。该项目包括四个协同研究重点。首先，联合文本和行为分析。为了利用尽可能多的人工生成数据类型并捕获它们之间的依赖关系，该项目开发了一套新颖的概率生成模型来执行文本和行为数据的综合分析。第二，基于任务的在线决策优化。传统的静态、临时和被动的人机交互不足以优化人类的动态决策过程。为了解决这一限制，用户的纵向信息搜索活动被组织成任务，其中应用新的在线学习算法来主动推断用户的意图，并使系统适应长期效用优化。第三，可解释的个性化。现有的个性化系统对其用户来说是黑盒子。用户通常无法控制他们的信息如何用于个性化系统。为了帮助普通用户了解系统的行为是如何定制的，并增加他们对此类系统的信任，构建了统计学习算法来生成面向系统和面向用户的解释。第四，系统实现与原型设计。用户研究是在一个原型系统中进行的，该原型系统集成了本项目开发的所有算法，以评估部署的算法。来自真实用户的评估和反馈被循环，以完善所开发算法的假设和设计。该项目的预期结果包括：1)开源工具和web服务，将提供对各种应用程序（如搜索日志、论坛讨论和意见评论）中人工生成的文本数据和行为数据的联合分析；2)标注的语料库和新的评价指标，使研究者能够在相关领域进行后续研究。

Humans are both producers and consumers of Big Data. Modeling the role of humans in the process of mining Big Data is thus critical for unleashing the vast potential of mined knowledge in various important domains such as health, education, security, and scientific discovery. The objective of this research is to build a human-centric learning framework, which harnesses the power of human-generated Big Data with computational models. This research project also focuses on incorporating conducted research activities into teaching materials for student training and education in the areas of information retrieval and data mining. In addition, specialized training in data analytics methods and applications is provided to enhance the education of Big Data related techniques for non-STEM and community college students.This project comprises four synergistic research thrusts. First, joint text and behavior analysis. To exploit as many types of human-generated data as possible and capture the dependencies among them, this project develops a set of novel probabilistic generative models to perform integrative analysis of text and behavior data. Second, task-based online decision optimization. Traditional static, ad-hoc and passive machine-human interactions are inadequate to optimize humans' dynamic decision making processes. To address this limitation, users' longitudinal information seeking activities are organized into tasks, where new online learning algorithms are applied to proactively infer users' intents and adapt the systems for long-term utility optimization. Third, explainable personalization. Existing personalized systems are black boxes to their users. Users typically have little control over how their information is used to personalize systems. To help ordinary users be aware of how the system's behavior is customized and increase their trust in such systems, statistical learning algorithms are built to generate both system-oriented and user-oriented explanations. Fourth, system implementation and prototyping. User studies are conducted in a prototype system integrated with all the algorithms developed in this project to evaluate the deployed algorithms. Evaluation and feedback from real users are circulated back to refine the assumptions and design of the developed algorithms. Expected results of the project include: 1) open source tools and web services that will provide joint analysis of human-generated text data and behavior data in various applications, such as search logs, forum discussions, and opinionated reviews; and 2) annotated corpora and new evaluation metrics that will enable researchers to conduct follow-up research in related domains.