LSTM Cluster: An Integrated Approach to Cluster Students' Problem Solving Sequences in Log Files
Qi Guo · ERA: Education and Research Archive (University of Alberta) · 2018
Modern technology-based assessments have the capacity to record every student-computer interaction in log files. Cluster analysis of log files could yield insights about students’ problem solving strategies and their misconceptions. However, current cluster analysis algorithms often rely on expert-selected features and they do not take the order of student actions into account. To address these limitations, this study proposes a novel deep learning approach to cluster student actions in log files. The proposed method, Long Short Term Memory (LSTM) cluster, first extracts features relevant to the problem from sequential data, and then clusters students based on the extracted features. In order to demonstrate and evaluate LSTM cluster, a real data study and a simulation study were conducted. In the real data study, LSTM cluster identified four different problem-solving strategies. A detailed examination of these strategies suggested that the order of student actions was indeed important. The simulation study suggested that LSTM cluster had good accuracy for large sample sizes. However, when sample size is small, the number of student actions needs to be reduced in order to prevent overfitting.