A Music Recommendation Method for Large-Scale Music Library on a Heterogeneous Platform

摘要

Currently, music recommendation system is a research focus in music information retrieval and a typical system can handle millions of music in real time. However, online music libraries have exceeded ten-million magnitudes, such as Amazon MP3, which results in mismatching between music recommendation systems and music libraries. Thus, this paper presents a music recommendation method for retrieving the large-scale music library on a heterogeneous platform. Based on the music similarity algorithm, by combining the indexing mechanism with GPU hardware acceleration, we further enhance the processing scale of the proposed method. Experiments show that, without lowering the retrieval accuracy, the proposed music recommendation method has the ability to handle ten-million magnitude libraries online in a single server.