Christian Doppler Lab for Recommender Systems

This research initiative aims to address the growing concerns surrounding recommender systems by developing more nuanced approaches that consider domain-specific contexts and user preferences. Beyond traditional accuracy metrics, the focus is on incorporating measures of novelty, diversity, and serendipity in recommendations. A multi-level user model will be introduced, encompassing individual, group, and network levels, utilizing multi-faceted feature sets to capture implicit preferences. Clustering techniques and social relations will further enhance user understanding. The goal is to create a more comprehensive user model facilitating systematic exploration of beyond-accuracy measures across various domains. Additionally, the research will analyze the dynamics of these measures over time and evaluate the long-term impacts of bias. This endeavor aligns with the principles of Digital Humanism, emphasizing fairness and broader societal concerns in algorithmic systems.