Persong: Multi-Modality Driven Music Recommendation System

Abstract

In this work, we develop PerSong, a music recommendation system that can recommend personalised songs based on the user’s current status. First, multi-modal physiological signals, namely visual and heart rate, are collected and combined to construct multi-level temporal sequences. Then, we propose a Global-Local Similarity Function (GLSF) based music recommendation algorithm to establish a mapping between the user’s current state and the music. Our demonstrations have attended a quite number of exhibitions and shown remarkable performance under diverse circumstances. We have made the core of our work publicly available: https://github.com/yrz7991/GLSF/tree/master.