SBIR Phase I: Creating the Critical University-to-Workforce Connection: Providing Quantitative Evidence of Collaboration Skills

该SBIR第一阶段项目将开发新的方法，用于培训和评估在协作环境中提供和接收建设性反馈的技能。尽管公司将协作技能列为新员工最重要的技能之一，但美国的高等教育系统却未能对其进行培训。过去的研究已经确定了最有效地鼓励同伴进步的特定类型的反馈，但今天的高等教育教师没有现成的方法来有效地评估学生使用这些同伴反馈策略的能力。该项目将开发简单易用的软件工具，以帮助教师培训同行反馈技能，以及自动评估学生提供和接收建设性反馈的能力的新方法。通过将当今雇主最看重的协作技能培训民主化，该项目旨在减少大学毕业生的未充分就业，并通过建立一支训练有素、生产率更高的劳动力队伍来增加美国的经济产出和税收。该项目将开发一个创新的同行评估工具，并结合新颖的自然语言处理方法，产生对学生基于书面反馈给予和接收建设性反馈的能力的准确衡量。自然语言处理方法将结合用于高效语言处理的单词向量、现有研究文献中概述的用于句子分类的可概括神经网络结构、以及基于现有同行反馈研究的手动定义的特征。研究工作将集中于从合作的高等教育教室收集大型训练数据集，并迭代完善深度学习自然语言处理算法，以实现90%的目标准确率。自动化这一评估过程将允许对学生协作能力进行广泛和持续的评估，而不需要对美国教师队伍进行昂贵和耗时的再培训。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。

This SBIR Phase I project will develop new methods for training and assessing the skill of providing and receiving constructive feedback in collaborative environments. The United States higher education system is failing to train collaboration skills, despite companies listing it as one of the most important skills for new hires to have. Past research has identified specific types of feedback that most effectively encourage peers to improve, but today's higher education instructors have no ready means to effectively assess students' abilities to use these peer feedback strategies. This project will develop simple-to-use software tools to help instructors train peer feedback skills and new methods for automatically assessing student ability to provide and receive constructive feedback. By democratizing the training of the collaborative skills most sought after by today's employers, this project aims to reduce underemployment of college graduates and increase the economic output and tax revenues of the United States by building a better trained and more productive workforce.This project will develop an innovative peer-assessment tool coupled with novel natural language processing methods producing accurate measurements of student ability to give and receive constructive feedback based on written feedback. The natural language processing methods will combine word vectors for efficient language processing, a generalizable neural net structure for sentence categorization outlined in existing research literature, and manually defined features based on existing peer feedback research. Research efforts will focus on gathering a large training dataset from partnering higher education classrooms, and iteratively refining a deep-learning natural language processing algorithm to achieve a target 90% accuracy. Automating this assessment process will allow broad and continuous assessment of student collaborative ability without requiring expensive and time-consuming retraining of the United States instructor workforce.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.