BIGDATA: IA: DKA: Collaborative Research: Learning Data Analytics: Providing Actionable Insights to Increase College Student Success

15年来，六年高等教育毕业率一直保持在59%左右；不到一半的大学毕业生在四年内完成学业。这有很高的人力、经济和社会成本。美国国家研究委员会（National Research Council）已经确定，迫切需要开发创新方法，以提高学生的保留率、毕业率和就业准备水平。该项目的目标是开发新的计算方法来分析大量不同类型的教育和学习数据，以帮助(a)为学生发现成功的学术途径；(b)改善教师的教学方法；(c)提高院校学生的持久性和留校率。项目成果旨在帮助学生选择适合他们需求、能力和学习风格的课程，并有可能（更快）毕业；帮助教师更好地满足学生的需求；并为顾问和机构提供提高保留率和持久性所需的分析。所提出的研究将产生新的动态系统建模、协同过滤和多任务学习方法。利用动态状态空间系统对学生的知识演变进行建模是一个关键的创新；提出的研究将开发用于等级预测的新型协同系统识别和协同卡尔曼滤波技术。技术创新包括用于不断发展的数据集的监督学习方法，如线性和非线性多任务学习以及具有潜在变量控制分组的协作多回归模型。这些创新将整合成三个试点应用程序：面向学生的DegreePlanner、面向教师的CourseInsights和面向学术顾问的StudentWatch。

The six-year higher-education graduation rate has been around 59% for over 15 years; less than half of college graduates finish within 4 years. This has high human, economic and societal costs. The National Research Council has identified a critical need to develop innovative approaches to improve student retention, graduation, and workforce-preparedness. The objective of this project is to develop new computational methods to analyze large and diverse types of education and learning data to help (a) discover successful academic pathways for students; (b) improve pedagogy for instructors; and (c) enhance student persistence and retention for institutions. The project outcomes are designed to help students select courses that fit their needs, capabilities, and learning styles, and are likely to lead to (faster) graduation; help instructors to better meet student needs; and give advisors and institutions the analytics needed to improve retention and persistence. The proposed research will produce new dynamical system modeling, collaborative filtering, and multi-task learning methods. Modeling the evolution of a student's knowledge using a dynamical state-space system is a key innovation; the proposed research will develop novel collaborative system identification and collaborative Kalman filtering techniques for grade prediction. Technical innovations include supervised learning approaches for evolving datasets, such as linear and non-linear multi-task learning and collaborative multi-regression models with controlled grouping of the latent variables. These innovations will coalesce into three pilot applications: DegreePlanner for students, CourseInsights for instructors, and StudentWatch for academic advisors.