Towards Personalized Explainability: The Development of a Modular Recommender System to Assess Explanation Suitability for Different Personas

This thesis presents the analysis and development of a modular Recommender System designed to assess the suitability of specific explanations for different Personas. The system leverages the analysis of three post-hoc explainers - Local Interpretable Model-Agnostic Explanations (LIME), High-Precision Model-Agnostic Explanations (Anchors), and LOcal Rule-based Explanations (LORE) — and incorporates a lightweight definition of Personas with specific preferences. The core idea is that users may have different needs or capabilities when interacting with explainers, and the system is designed to recommend the most suitable explanation for each user type based on predefined and evolving preferences, reflecting changing user behaviors or contexts. The system supports a personalized approach to explainability, moving beyond the limitations of a one-size-fits-all paradigm.