2023 ASEE Annual Conference & Exposition

A vast body of research demonstrated the learning benefits and importance of collaborative learning (i.e., pedagogical approaches involving students learning collectively by working on collaborative assessments that require joint intellectual effort). Such benefits include increasing student retention, higher-level thinking skills, improved communication skills, responsibility for students' own learning, and more. Acknowledging these benefits, more instructors are integrating collaborative learning into their curriculum (Järvenoja et al., 2020).

With this increased adoption of collaborative learning approaches, instructors must understand their students' problem-solving approaches during collaborative learning activities to better design their class activities. Among the multiple ways to reveal collaborative problem-solving processes, temporal submission patterns is one that is more scalable and generalizable in Computer Science education. In this paper, we provide a temporal analysis of a large dataset of students' submissions to collaborative learning assignments in an upper-level database course offered at a large public university. This research uses sequential pattern mining techniques to identify temporal patterns that capture students' problem-solving strategies when working on collaborative assignments. We study the log data collected from an online assessment and learning system, which contains the timestamp data of each student's submissions to a problem on the collaborative assignment. Each submission was labeled as quick (Q), medium (M), or slow (S) based on its duration and whether it was shorter or longer than the 25th and 75th percentile.

Our analysis revealed seven submission patterns after applying two compacting rules: Q, Q-LNG, M, M-LNG, QM-LNG, S, and S-LNG, where "LNG" dictates that the compacting rule was applied. The compacting rule indicates that there were at least three consecutive submissions of the same pattern (i.e., Q-LNG is a compaction of Q-Q-Q). We then investigated how submission patterns link to become discriminating sequences by computing the transition probabilities with L* metrics. We identified transition pairs that include the same submission patterns in different transition positions (i.e., transit from and transit to) to construct more extended transition patterns by combining transition pairs. For example, the transition pairs M-LNG to S-LNG and S-LNG to Q have a correlation of 0.782 (p<0.001), so we can generate a three-pattern transition: M-LNG to S-LNG to Q. With these transition patterns, we can then better pinpoint collaborative learning problem-solving behaviors among students.