Modeling Multimedia Ad Exposure: The Role of Low-Level Audiovisual Features

This study investigates whether low-level audio and video features can explain the variance in multimedia exposure as measured by the Multimedia Advertising Exposure Scale (MMAES). The goal is to understand the role of these features in predicting exposure and explore whether incorporating nonlinear relationships based on interactions can improve modeling. An observational study with young participants (N=287) evaluated exposure to eight video advertisements. Linear and polynomial regression models were used to predict MMAES scores using low-level features. Results indicated that polynomial models outperformed linear models, capturing complex interactions and providing better predictive accuracy. The best polynomial model explained 34.6% of the variance in MMAES scores, with an MAE of 0.207 and an RMSE of 0.269. Bootstrap analysis confirmed model robustness, showing lower error rates and stable coefficients for polynomial models. SHapley Additive exPlanations (SHAP) analysis highlighted key features and their interactions, improving interpretability. These findings underscore the importance of nonlinear relationships in modeling multimedia exposure, with implications for optimizing multimedia advertising strategies.