Emotional Arousal Effects on Collective Memory: Machine Learning-Based ECG Analysis in Interactive Installations

Collective memory, serving as a crucial mechanism for maintaining group identity and social cohesion, has found new pathways of activation through interactive installations and public art in recent years. However, existing research relies predominantly on qualitative analysis, lacking systematic quantitative verification of the emotional arousal process. This study recruited 58 participants to conduct experiments based on two collective memory interactive installations. Data were collected using the PANAS-SF mental scale and electrocardiogram (ECG) measurements at three stages: baseline (T0), post-experience (T1), and one-week follow-up (T2). K-Nearest Neighbors (a supervised machine learning model) was employed to identify and classify relationships between ECG signals and emotional labels. Repeated measures ANOVA was used to analyze the psychological scale data. The results indicate that the KNN model achieved high recognition accuracy for high-arousal emotions (training set recall: 85.1%, precision: 88.9%). The psychological scales revealed a significant increase in positive emotions at the T1 stage—with the sense of calmness rising from 17 to 32—which was partially sustained into the T2 stage, whereas negative emotions showed little change. Interview feedback further highlighted the critical role of auditory cues and operational interaction in enhancing immersion and emotional resonance. These findings revealed that collective memory installations can not only effectively stimulate positive emotions in the short term but also sustain a certain level of calm experience. The incorporation of multimodal methods offers a new framework and practical insights for emotion evaluation in interactive experiences.