| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 107 === The main goal of a music recommendation system is to provide music items that people may like. However, when making music recommendations, it is limited by the sparsity problem due to the large scale of music items and users. Besides, the sparsity problem also becomes more serious because of the lack of music explicit feedback. Therefore, researchers have been working hard to find new solutions to solve the above-mentioned problems and try to improve the accuracy of the music recommender system. This study uses the implicit feedback of music domain and implements the data transformation to transform the data type from the implicit nature to the explicit data. After that, a collaborative filtering method is adopted due to the high efficiency in music recommendation. Finally, the hetrec2011-lastfm-2k music dataset is used as our experimental dataset. In order to solve the sparsity problem and enhance the accuracy of music recommendation, this study tries to combine singular value decomposition with cluster-based collaborative filtering as the main recommendation technique. In addition to use Singular Value Decomposition (SVD), cluster-based collaborative filtering can not only remove the noisy data but also find the better preference group. After that, the system will generate a preference model to find the predicted music preference of users. Finally, the proposed method in this thesis can improve the cluster effect on large music data and enhance the prediction accuracy of the music recommender system.
|
|---|
