Pilot Study on Applying Natural Language Processing Techniques to Classify Student Responses to Open-Ended Problems to Improve Peer Review Assignments

As an educational tool, peer review can be a valuable way to provide students feedback without a significant increase in instructor workload. Despite all that is currently known about our students, the best mechanism for assigning reviewers to reviewees in a peer review of artifacts is still considered to be blind, random assignment. The underlying conjecture of this research project is that “there has to be a better way”. This paper represents a follow-up to earlier work by the author [1]. That study presented the results of an attempt to develop a classification schema using a large archival database of student work. This paper takes the resulting algorithms produced from that archival dataset and applies them to new student work, identifying how well the archival-based classification works on a new data set. The implications of that application on future algorithm design will be discussed as well as the next steps for the research.