(Em)Powering Recommendations: Analyzing the Energy Efficiency of Recommender System Algorithms

Recommender systems provide personalized content to users, from movie and music recommendations to specific products or travel itineraries. While these systems have a substantial impact on user experience and business performance, they also require significant computational resources. With the rise of deep neural networks also in the context of recommender systems, concerns about their energy consumption and environmental impact have also increased. This thesis investigates the energy efficiency of different recommender system algorithms, focusing on traditional approaches (e.g., collaborative filtering, matrix factorization) and state-of-the-art machine learning techniques. Through empirical analysis, the goal is to measure and compare the energy consumption of these algorithms in various deployment scenarios. The study aims to identify energy-efficient alternatives and algorithm improvements while maintaining system accuracy.