Building an effective recommender system for Audio-on-Demand platforms presents unique challenges.Beyond the user-item interaction data, the inclusion of user and item metadata holds promise for enhancing recommendation quality. However, when faced with limited availability of metadata, can we still construct a robust recommender? This thought-provoking talk delves into the realm of recommender systems, exploring how simple collaborative filtering techniques can be adapted to accommodate scarce metadata. Furthermore, we explore an intriguing alternative approach: treating user-item interactions as interconnected nodes within a graph. By leveraging graph-based methodologies, we unveil a powerful graph-based recommender system. Through a captivating use-case, we highlight the potential of this novel viewpoint, elucidating its ability to uncover hidden patterns and enhance recommendation accuracy.
Read more at the Link
ML Con Berlin 2023, Dr. Mirza Klimenta
