An Online Music Recommendation System Based on Annotations about Listeners' Preference and Situation

Katsuhiko KAJI
Graduate School of Information Science, Nagoya University

NTT Communication Science Laboratories
Katashi NAGAO
EcoTopia Science Institute, Nagoya University


Automatic playlist generation is one of the powerfulintermediaries for music providing services. Many people can be promotedto music contents.In this paper, we propose a playlist generation scheme. By using lyricsand annotation, musical similarity and user's similarity is found. Thenit generates a playlist according to a user's preference and situation.Additionally, the user can provide a feedback such as a favorite music andwhether the playlist is recommendable. The system transform thecharacteristic values concerning the user's preference and situation sothat it adapts to each user.

1 Introduction

2 Playlist Recommendation System

System Architecture

Fugure1: System Architecture

Triple feature space

Fugure2: Triple feature space

2.1 Estimation of similarity between musical piece using lyrics and annotation


2.2 Collaborative filtering

Overview of collaborative filtering

Fugure4: Overview of collaborative filtering



2.3 Transcoding


2.4 Interaction

An example of generated playlist

Fugure8: An example of generated playlist


3 Pilot Study

The result of experimentation

Fugure10: The result of experimentation

4 Future Work

5 Summary

6 Acknowledgement