EAGER: Technology to Review Online Videos for Education (TROVE)

在线视频正变得越来越受儿童欢迎。这对家长和教育工作者来说是一个挑战，他们希望他们观看教育视频，但可能缺乏能力或时间在快速增长的在线视频宇宙中区分教育内容和非教育内容。据我们所知，目前还没有机器学习方法来对教育视频内容进行分类；目前的方法依赖于人类来识别视频内容。但是，当平均每分钟有大约500小时的内容上传到YouTube时，人工内容审查就跟不上步伐。这个项目的目标是开发用于教育的在线视频审查技术(TROVE)，这是一个基于机器学习的工具，可以在大量视频中识别早期儿童数学内容。这一能力将使新的方法能够增加幼儿接触视频中适合发展的数学内容，这已被证明可以改善数学学习结果。Trove将为识别视频中的一系列主题奠定基础，包括识字、科学和社会情感内容。此外，在该项目下开发的技术进步将在其他领域得到应用，包括自适应学习、社交媒体分析、宣传检测和视频摘要。我们由教育和机器学习研究人员组成的多学科团队将开发一个内容分类引擎，以识别在线视频中的数学内容。我们将根据Head Start早期学习成果框架和共同核心国家标准，在幼儿、学龄前和幼儿园水平上定义适合发展的数学内容。为了训练内容分类引擎，已经证明了识别目标数学内容的可靠性的研究人员将在100个小时的在线视频中对数学内容进行注释。该项目的核心研究问题是，与人类相比，内容分类引擎能在多大程度上准确识别视频中的幼儿数学内容？为了回答这个问题，我们将把Trove识别的数学内容与研究人员在一组未用于训练分类引擎的视频中识别的数学内容进行比较。我们将通过同行评议的期刊文章和博客文章与教育技术研究人员和开发人员、教育工作者和政策制定者分享我们的发现。TRUVE具有通过接触高质量、适合发展的教育视频来支持幼儿学习的变革性潜力。此外，这项技术可能会对儿童接触教育和非教育视频内容的影响进行大规模研究。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Online videos are becoming increasingly popular with young children. This presents a challenge for parents and educators who want them to watch educational videos but may lack the ability or time to distinguish educational from non-educational content within the rapidly growing universe of online video. To our knowledge, there are currently no machine learning methods for classifying educational video content; current methods rely on humans to identify video content. But human content reviews cannot keep pace when, on average, approximately 500 hours of content are uploaded to YouTube every minute. The goal of this project is to develop Technology to Review Online Videos for Education (TROVE), a machine learning-based tool to identify early childhood mathematics content in a high volume of videos. This capability will enable new approaches to increase young children’s exposure to developmentally appropriate mathematics content in videos, which has been shown to improve mathematics learning outcomes. TROVE will lay the groundwork to identify a range of subjects in videos, including literacy, science, and social-emotional content. Further, the technological advances developed under this project will have applications in other fields, including adaptive learning, social media analytics, propaganda detection, and video summarization. Our multidisciplinary team of education and machine learning researchers will develop a content classification engine to identify mathematics content in online videos. We will define developmentally appropriate mathematics content at the toddler, preschool, and kindergarten levels based on the Head Start Early Learning Outcomes Framework and Common Core State Standards. To train the content classification engine, researchers who have demonstrated reliability in identifying the target mathematics content will annotate the mathematics content in 100 hours of online videos. The project’s central research question asks, how accurately can the content classification engine identify early childhood mathematics content in videos, as compared to humans? To answer this, we will compare the mathematics content identified by TROVE to that identified by researchers in a set of videos that were not used to train the classification engine. We will share our findings with education technology researchers and developers, educators, and policymakers via a peer-reviewed journal article and blog post. TROVE has transformative potential to support young children’s learning through exposure to high-quality, developmentally appropriate educational videos. Further, this technology may enable large-scale research on the impacts of children’s exposure to educational and non-educational video content.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.