Exploring a Flexible Computational Method for Comparing Massive Interaction Data from Science Visualizations

Undergraduates: Sweta Karlekar, Emily Toutkoushian

Faculty Advisor: Kelly Ryoo
Department: Computer Science

---

Interactive visualization technologies, such as simulations, can automatically collect massive data on how students interact with tools while learning science. Grouping and comparing such data can identify students¿¿¿ interaction patterns which can be used to create more personalized learning environments. Existing methods for analyzing and comparing interaction data are often inflexible and hard to incorporate in newly developed visualizations. The purpose of this study is to explore the effects of a new computational method that uses numerical encoding and Levenshtein edit distance, developed to be simple and flexible for encoding and comparing student interaction data. The study involved linguistically diverse 8th-grade students from a low-income middle school. Students engaged in predict-observe-reflect activities using interactive simulations to explore chemistry concepts. All students¿¿¿ interactions with the simulations were automatically logged, including the order of various button clicks. Reflection questions were scored based on a rubric evaluating content, claim, and evidence provided by the student. Students were grouped based on reflection score and, upon processing the interaction data with the proposed method, statistically significant differences were found between mean edit distances of each score group. The results suggest that this new method shows promise for comparing interaction data as differences in learning patterns were found between the score groups.