Recommendation systems are widely used in music streaming services. This paper presents a system to collect user's music preference and music textual features in YouTube as well as to provide music recommendations based on collaborative filtering. As cold start and data sparsity are two severe issues in collaborative filtering, additional features for the item are necessary. We propose a method to aggregate both implicit feedback and textual features collected from YouTube to improve the recommendation performance. Experiment results indicate that the recommendation playlists generated by this system both match individual's and group preference.