CAREER: Time-Aware Multi-Objective Recommendation in Online Learning Environments

在线教育在劳动力培训、技能发展和终身学习方面发挥着越来越重要的作用。考虑到在线学习系统的规模及其对学生自我调节的依赖，自动推荐和教学工具对学生在在线教育中取得成功至关重要。理想情况下，这些工具应该为学生提供个性化的指导，在适当的时间使用最有效的学习材料（例如，要解决的问题或要观看的视频讲座），以有效地完成他们的个人学习目标（例如，满足他们的兴趣或在最短的时间内学习一个主题）。然而，目前的解决方案侧重于指导一种学习活动，忽视了学习时间间隔的重要性，只满足所有学生的一个目标。该项目将研究新一代的教育推荐系统，通过自动检测和平衡学生不同的、可能相互冲突的学习目标，实现学生的长期目标。该项目将开发计算模型和算法，为学生提供各种类型的个性化学习活动和最佳选择的学习时间。由此产生的解决方案将适用于机器学习和数据挖掘领域，特别是健康和健身等领域的长期效用和时间敏感的推荐系统。这项研究的成果将提高在线教育的可访问性，更好地服务于代表性不足的学习者。研究结果可用于教育领域，以改善学生的学习。该项目包括一个综合教学计划，以促进培养下一代计算机科学与教育领域融合的跨学科本科生和研究生。本项目旨在研究时效性强、多目标、多类型、个性化的教育推荐模型和算法。本项目设计的推荐系统将根据学生的长期学习兴趣、行为偏好和学习目标进行优化。他们将能够向学生推荐评估和非评估类型的学习材料，并为他们建模最佳学习时间间隔。该项目的贡献是通过三个研究重点来实现的。在第一部分中，设计并集成了个性化知识跟踪模型、学生选择模型和行为偏好模型，以模拟促进多目标推荐的长期奖励。在第二个重点中，该项目将研究多种类型的知识模型和算法，在考虑学生的多重目标的同时，向学生建议评估（例如，问题）和非评估（例如，视频讲座）的学习材料。第三个重点是教育推荐系统的新模式，通过结合点过程建模来检测不同活动之间的连续时间影响，并找到最佳的个性化返回学习时间。这些研究重点建立在教育数据挖掘、推荐系统、时间过程建模和强化学习文献的基础上，并扩展了它们，同时为学生的学习过程提供了可操作的见解。为了评估该项目的结果，我们制定了一个综合评估计划，包括模拟、对真实世界数据集的离线评估、与实时系统的在线集成以及用户研究。该项目的成果将通过开源软件和出版物传播到更广泛的机器学习、人工智能和教育数据挖掘社区。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Online education is playing an increasingly essential role in workforce training, skill development, and life-long learning. Given the scale of online learning systems and their dependence on student self-regulation, automatic recommendation and instructional tools are crucial for students' success in online education. Ideally, these tools should provide personalized guidance to students to work with the most effective type of learning material (e.g., a problem to solve or a video lecture to watch), at the right time, to efficiently accomplish their personal study goals (e.g., to fulfill their interests or to learn a topic in the shortest time). Current solutions, however, focus on instructing one type of learning activity, ignoring the importance of studying time intervals, and only satisfying one goal for all students. This project will research a new generation of educational recommender systems towards achieving students’ long-term goals by automatically detecting and balancing between their different, potentially conflicting study goals. This project will develop computational models and algorithms that can suggest various types of personalized learning activities and optimally selected study times for students. The resulting solutions will be applicable to machine learning and data mining fields, especially, to long-term utility and time-sensitive recommender systems in domains such as health and fitness. The products of this research will improve the accessibility of online education to better serve underrepresented learners. The findings can be used in the Education domain to improve students' learning. This project includes an integrated teaching plan that facilitates training the next generation of interdisciplinary undergraduate and graduate students in the convergence of Computer Science and Education fields. This project aims to research time-aware, multi-objective, multi-type, personalized educational recommender models and algorithms. The recommender systems designed in this project will be optimizing for students’ long-term learning interests, behavioral preferences, and learning goals. They will be capable of recommending both assessed and non-assessed types of learning materials to students and modeling the best study time intervals for them. The contributions of this project are realized via three research thrusts. In the first thrust, personalized knowledge tracing models, student choice models, and behavioral preference models are designed and integrated to model long-term rewards that facilitate multi-objective recommendations. In the second thrust, the project will investigate multi-type knowledge models and algorithms to suggest both assessed (e.g., problems) and non-assessed (e.g., video lectures) learning materials to students while considering their multiple objectives. The third thrust focuses on a new modality of educational recommender systems by incorporating point process modeling to detect the continuous-time influence between different activities and find the optimal personalized return-to-study time. These research thrusts build upon educational data mining, recommender systems, temporal process modeling, and reinforcement learning literature and extend each of them while contributing actionable insights in student learning processes. To assess the results of this project, a comprehensive evaluation plan is incorporated that includes simulations, offline evaluation on real-world datasets, online integration with live systems, and user studies. The results of this project will be disseminated to the broader Machine Learning, Artificial Intelligence, and Educational Data Mining communities via open-source software and publications.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.