Personalized Feedback in Case-Method Study

Abstract

The long-term goal of our research is to create a case method learning without teachers. To achieve this goal, we found that personalized feedback was effective for students in a preliminary experiment. Then, we investigated the possibility of evaluating description problem answers by machine learning. As a result, our study showed that we could score them automatically on high-accuracy utilizing LSTM with attention. The accuracy would become higher utilizing posterior probability in the network, and we found that the visualization of attention was quite moderate. Moreover, we examined the number of answers to keep the accuracy high.