Collaborative Research: Modeling Social Interaction and Performance in STEM Learning

这个教育与学习研究（REAL）项目源于2014年10月的一个关于数据密集型研究以改善教与学的想法实验室。这一努力的目的是：(1)将来自不同学科的研究人员聚集在一起，培养新颖的、变革性的、多学科的方法，利用大型教育相关数据集中的数据，创造可操作的知识，以改善中期的STEM教学和学习环境；(2)从长远来看，彻底改变学习方式。在这个项目中，来自教育测试服务中心、哥伦比亚大学师范学院、亚利桑那州立大学和北卡罗莱纳州立大学的研究人员将进行数据驱动的探索性分析，以确定在特定学习环境中社会互动影响学习成果的关键地方，目标是改善大规模STEM课程的教与学。这项研究利用了大规模混合和在线学习环境（包括大规模开放在线课程，简称MOOCs）中留下的数据痕迹。研究人员将在这些课程的背景下开发一个全面的社会学习模型，该模型将能够评估课堂背景下个人的合作需求，以及他们正在进行的合作的质量。这样的诊断将允许教师和自动化系统向学习者提供关于他们可能与之合作的同伴的建议，以提高他们的学习（例如，关于能够促进更好理解和发展重要学科能力的社会互动的种类）。一个跨学科的研究团队，拥有理论驱动的教育数据挖掘、自然语言处理、心理测量学、社会网络分析和协作学习的计算机支持方面的专业知识，将合作探索混合式和在线课程中的学习者何时从社会互动中受益，并了解如何识别更多和更少的生产性协作互动。研究人员将使用来自三个混合和在线课程的数据（例如，记录协作讨论的日志文件，围绕良好仪器示例的个人和协作交互，同伴辅导课程，结对编程实验室，结对项目）和各种数据分析方法（例如，文本分析，机器学习）来确定：(1)可以从可用数据中识别需求和交互的认知，社会和情感维度；(2)哪些分析有助于提供面向行动的协作建议；(3)提出这些建议可能需要哪些其他类型的数据。这一探索将以学习的社会互动理论为基础（例如，自我解释，与自己和他人的辩证法，最近发展区，班杜拉的社会学习理论，外围和向心参与）。

This Research on Education and Learning (REAL) project arises from an October 2014 Ideas Lab on Data-intensive Research to Improve Teaching and Learning. The intentions of that effort were to: (1) bring together researchers from across disciplines to foster novel, transformative, multidisciplinary approaches to using the data in large education-related data sets to create actionable knowledge for improving STEM teaching and learning environments in the medium term; and (2) revolutionize learning in the longer term. In this project, researchers from the Educational Testing Service, Columbia University Teachers' College, Arizona State University, and North Carolina State University will conduct data-driven, exploratory analyses to identify key places where social interactions impact learning outcomes in specific learning environments, with the goal of improving teaching and learning in large-scale STEM courses. This research takes advantage of data traces left in large-scale blended and online learning environments (including massively open online courses, or MOOCs). The researchers will develop a comprehensive model for social learning in the context of such courses that will enable assessment of both the collaborative needs of individuals within the context of a class, and the quality of collaborations they are carrying out. Such diagnoses will allow both instructors and automated systems to provide advice to learners about the peers they might work with to enhance their learning (e.g., regarding the kinds of social interactions that will foster better understanding and development of important disciplinary capabilities). An interdisciplinary team of investigators with expertise in theory-driven educational data mining, natural-language processing, psychometrics, social-network analysis, and computer support for collaborative learning will collaborate to explore when learners in blended and online classes benefit from social interactions, and to understand how to identify more and less productive collaborative interactions. The researchers will use data from three blended and online classes (e.g., log files capturing collaborative discussions, individual and collaborative interactions around well-instrumented examples, peer tutoring sessions, pair programming labs, paired projects) and a variety of data analysis approaches (e.g., text analysis, machine learning) to determine: (1) which cognitive, social, and affective dimensions of need and interaction can be identified from available data; (2) which analyses are useful in providing action-oriented collaboration advice; and (3) what additional types of data may be needed for making such recommendations. This exploration will be grounded in theories of social interactions for learning (e.g., self-explanation, dialectic with oneself and others, zone of proximal development, social learning theory of Bandura, peripheral and centripetal participation).