Physiological Signals and Affect as Predictors of Advertising Engagement

Area of Science:

• Human-Computer Interaction
• Affective Computing
• Biometric Signal Processing

Background:

• Understanding user engagement with advertising is crucial for marketing effectiveness.
• Physiological and affective signals offer objective measures of user responses.
• Previous research has explored various biosignals for engagement detection with mixed results.

Purpose of the Study:

• To investigate the efficacy of combining physiological signals (heart rate, electrodermal activity, pupil dilation, skin temperature) and affective states (valence, tiredness) for classifying advertising engagement.
• To compare the performance of different signal combinations and machine learning models in predicting user engagement.
• To identify optimal feature sets for robust and efficient ad engagement modeling.

Main Methods:

• Collected ground truth data on ad engagement using the User Engagement Scale from 53 young adults.
• Utilized physiological signals: heart rate, electrodermal activity, pupil dilation, and skin temperature.
• Employed three gradient-boosting classifiers (LightGBM, HistGradientBoostingClassifier, XGBoost) with signal fusion techniques.

Main Results:

• Classifiers trained on a fusion of skin temperature, valence, and tiredness (5 features) outperformed those using all 30 signals.
• The optimized feature set achieved higher average AUC ROC scores (e.g., HGBC = 0.70 ± 0.11) compared to using all signals (e.g., HGBC = 0.64 ± 0.10).
• The signal fusion incorporating skin temperature demonstrated superior performance over other signal combinations.

Conclusions:

• Specific physiological signals, particularly skin temperature, are highly informative for modeling advertising engagement.
• Signal fusion of selected affective and physiological data enhances predictive accuracy while simplifying the feature set.
• This approach offers a more effective and efficient method for quantifying user engagement with digital advertisements.