•  
  •  
 

Abstract

Behavioral targeting is a key part of the modern advertising web's algorithmic engine. However, it is unclear whether optimization processes worsen bias, promote unchecked spread in filter bubbles or lower overall users' trust levels. This paper introduces HARMONIA (Holistic Adaptive Regulatory Model for Optimizing Non-transparent Intelligent Advertising), a comprehensive, data-driven Explainable Artificial Intelligence (XAI) framework aimed at transforming behavioral targeting via transparency, interpretability, and adaptive ethical regulation. This paper conducted a comprehensive Explorative Data Analysis (EDA) on the public Criteo Display Advertising Dataset, which contains over 45 million records, to identify patterns in high-dimensional user-ad interaction space. This analysis uncovered latent behavioral signals that affect the relevance of ads based on users' online behavior. The analysis identified four interrelated behavioral dynamics: ad fatigue attenuation, diurnal engagement oscillations, device-driven preference divergence, and category-affinity dominance. These dynamics served as the foundational architectural principles for HARMONIA's design. The method uses gradient boosted prediction models and a multilayer explainability stack that includes SHAP for global interpretability, LIME for local surrogate approximation, and counterfactual reasoning for causal transparency. Quantitative evaluation indicates that HARMONIA maintains relevance accuracy (approximately 1.2% CTR), achieves a 31% enhancement in transparency metrics, and a 27% improvement in user-trust indices, while concurrently reducing systemic entropy by nearly one-third. This research redefines personalization to be self-explanatory and ethically sound AI by incorporating explainability as a regulatory mechanism in the adaptive ecosystem of complex digital advertising. This system takes explainable computational marketing from an idea to a full-scale implementation.

Download

Share

COinS