Interaction Density: Simple Press Logs Predict Exhibit Type with 0.778 AUC

A new study from researchers at a Bengaluru science experience centre demonstrates that minimal, privacy-preserving interaction logs can classify exhibit types accurately. Analyzing 2,816 visitor sessions across eight exhibits at two venues, the team derived 'interaction density' — presses per second — from just three default log fields: session start time, end time, and press count. The metric cleanly separates fast-paced games from deliberate quizzes (Mann-Whitney r=0.556, AUC=0.778). Intriguingly, visitors spent more time on games than quizzes (r=0.172), reversing the expected intensity-duration trade-off due to open-ended re-engagement loops in games with escalating difficulty. Raw press count alone explained far more dwell time variance (R²=0.527) than density (R²=0.081), but combining both improved prediction (R²=0.667). The study also validates its findings across venues and highlights session-length censoring artifacts.

The broader implication is methodological: any science centre with touch-enabled exhibits can conduct rigorous, falsifiable behavioral research using existing logs, bypassing costly sensors like eye-trackers or wearables. This opens the door for large-scale, privacy-safe studies on visitor engagement patterns. The work is presented in a 7-page paper on arXiv (cs.HC) with the raw dataset included as an ancillary file.