Automatic playlist generation by applying tabu search
- PDF / 1,751,982 Bytes
- 16 Pages / 595.276 x 790.866 pts Page_size
- 59 Downloads / 202 Views
ORIGINAL ARTICLE
Automatic playlist generation by applying tabu search Jia-Lien Hsu • Ya-Chao Lai
Received: 27 February 2012 / Accepted: 31 January 2013 Springer-Verlag Berlin Heidelberg 2013
Abstract In this paper, we propose a solution to the problem of playlist generation. In order to capture user listening preference and recommend playlists, we maintain user profiles by keeping listening history. Then, we apply the sequential pattern mining algorithm with multiple minimum supports on user profiles to derive constraints. Given a set of derived constraints, we apply the tabu search to generate playlists which match constraints as much as possible. Finally, we implement our prototype and perform experiments to show the feasibility, efficiency, and effectiveness of our approach. Keywords Constraint-based playlist generation Tabu search Sequential pattern mining with multiple minimum supports
1 Introduction Finding user favorite or desired music within a large amount of music collections remains a promising but challenging task. In the past, a variety of emerging music services have been proposed, such as recommendation and playlist generation. For example, when using Windows Media Player and iTunes, we are proposing to make custom playlists by manually adding songs. Some music service providers, e.g. playlist.com, offer customized online playlists in which users create, maintain and share playlists manually. This can be a tedious task which
J.-L. Hsu (&) Y.-C. Lai Department of Computer Science and Information Engineering, Fu Jen Catholic University, New Taipei City, Taiwan, R.O.C e-mail: [email protected]
requires much attention to constantly update those playlists in order to keep current. In this paper, we propose a system which automatically generates playlists for users. First, we organize user profiles, which keep track of user music listening history, as an input for the constraint-derived process in the next step. We apply the sequential pattern mining with multiple minimum supports algorithm (MMS-PS) on user profiles to mine the frequent occurring subsequences and then derive constraints for capturing the user music preference. By applying the tabu search against the derived constraints, we are able to generate playlists which satisfy those derived constraints as much as possible. When compared to previous works on playlist generation, our method is by far more distinguished in that it has two features. First, in this process, user behavior patterns will be captured and modeled by constraints. Based on the listening sequences, we derive listening behavior constraints. Second, these constraints are specified in a relative manner, which are more flexible and well interpreted. The rest of paper is organized as follows: Sect. 1 features related work on playlists generation; Sect. 2 defines and identifies the playlist generation problem; Sect. 3 proposes our method of deriving constraints and generating playlists; Sect. 4 demonstrates the experimental results of our approach; and, Sect. 5 conc
Data Loading...
