Group Assignments for Project-Based Learning Using Natural Language Processing-A Feasibility Study

Abstract

Group learning is commonly used in a wide range of classes. However, effective methods used to form groups are not thoroughly understood. In this study, we explore a quantitative method for creating project teams based on student knowledge and interests expressed in project proposals. The proposals are encoded to vector representations, ensuring that closely related proposals yield similar vectors. During this step, two widely used natural language processing algorithms are used. The first algorithm is based solely on the frequency of words used in the text, while the other considers context information using a deep neural network. The similarity scores for the proposals generated by the two algorithms are compared with those generated by human evaluators. The proposed method was applied to a group of senior students in a capstone design course in South Korea based on their project proposals on autonomous cars written in Korean. The results indicate that the contextualized encoding scheme produces more human-like text similarity vectors compared to the word frequency-based encoding scheme. This discrepancy is discussed from a context information standpoint in this study.