Improving personalized recommendations for cold-start users on the NetEase Cloud Music Platform

Abstract

The rise of big data and rapid digitization in the digital media industry have made recommendation systems essential for delivering relevant, personalized content to users. In the music streaming sector, platforms like NetEase face the cold-start problem when recommending content to new users with minimal interaction data. To address this, we developed and compared various techniques, including a User Similarity-Based Recommendation Algorithm, a User Preference Elicitation Recommendation Algorithm, a DeCS-Inspired Recommendation Algorithm, and a Discriminative Frequent Itemsets model. Our findings show that the DeCS-Inspired model performs best in data-rich scenarios, while Demographic-Based methods excel in cold-start situations. To optimize performance, we propose a hybrid approach that combines Demographic-Based techniques for cold-starts and transitions to the DeCS-Inspired model as user data grows.