LLaMAC: low-cost biosignal sensor based large multimodal dataset for affective computing

Chang-Gyu Lee ¹, Joo Young Kim ²

⁰Culture Wellness R&D Group, Korea Institute of Science and Technology, Gyeonggi-do, Republic of Korea. cglee@kitech.re.kr.

Scientific Data +

|

December 4, 2025

View abstract on PubMed

Summary

The LLaMAC dataset uses biosignals and questionnaires to predict media success through emotion prediction. This approach enables analysis of emotions, liking, and familiarity from physiological data.

Area Of Science

Affective computing
Human-computer interaction
Media psychology

Background

Predicting media success is challenging.
Emotion plays a key role in media reception.
Biosignals offer objective measures of emotional states.

Purpose Of The Study

Introduce the LLaMAC dataset for emotion prediction in media.
Facilitate biosignal-based prediction of emotions and liking.
Enable correlation analysis between continuous and discrete emotional data.

Main Methods

Collected biosignals (EEG, GSR, PPG, SKT, RESP) and questionnaire data from over 100 participants.
Validated biosignals using statistical metrics and signal-to-noise ratios.
Utilized Light Gradient Boosting Machine (LightGBM) for emotion classification.

Main Results

Demonstrated the feasibility of predicting emotions and liking from biosignals.
Established correlations between continuous (valence, arousal, dominance) and discrete emotional states.
Identified differences in biosignals related to media familiarity.

Conclusions

The LLaMAC dataset supports biosignal-based emotion and liking prediction.
Findings advance understanding of the relationship between physiological responses and media experience.
The dataset facilitates further research into affective computing and media engagement.