Development of Social Networks based on Music Recommendation Techniques for Assisting Active Music Listening

本研究旨在开发一种技术，能够以较低的计算成本向用户准确推荐音乐作品，并使用户能够通过人机交互来主动控制推荐结果。首先，我们研究了音乐内容的自动识别，可以用来提出建议。因此，我们成功地开发了一种从复调音乐音频信号中准确检测多个基频的方法，并显著提高了推荐的准确性。其次，我们一直致力于开发社交网络功能，使用户在进行推荐时能够有选择地专注于音乐内容，并发现其他有类似品味的用户。我们实现并检验了一个基于Web的推荐系统。未来的工作包括社交网络的完善和推荐系统的实际运行。

This research aims to develop techniques that can accurately recommend musical pieces to users at a low computational cost and enable users to actively control recommendation results through human-machine interaction. First, we have studied on automatic recognition of musical contents, which could be used for making recommendations. Consequently, we successfully developed a method that can accurately detect multiple fundamental frequencies from polyphonic musical audio signals, and significantly improved the accuracy of recommendations. Second, we have been working on development of social-network functions that enable users to selectively focus on musical contents in making recommendations and to discover other users who have the similar tastes. We implemented and examined a Web-based recommender system. Future work includes improvement of social networks and practical operations of the recommender system.