Automatic Playlist Generation
نویسندگان
چکیده
Digital music applications have become an increasingly popular means of listening to music. Applications such as Spotify allows the user to add songs to his/her playlist without downloading the song to his/her computer. The user can also be recommended songs by Spotify through Spotify’s “Discover" option. Pandora— an online radio—generates a radio station based on a single user-inputted artist, genre, or composer. For these types of applications, applying algorithms to learn user preferences is extremely important. In this report, we explore two different methods to generate automatic playlists based on user-input: 1. Gaussian Process Regression (GPR): This method takes in a set of seed songs (which can contain as little as a single song) inputted by the user to train a preference function that predicts the user preference for a new song to be considered for the playlist. 2. SVM + HMM: For this method, we assume that the user has generated a large number of seed songs (i.e. a user has liked hundreds of songs are pandora throughout the course of a year). For this method we also require a set of low-preference songs (i.e. the user has skipped hundreds of songs on pandora over a year). With a large training set of labelled data, we can apply classification algorithms such as SVM to determine if a new song will be liked or disliked by the user. Because we believe timbre to be an important predictor for music, we combine the SVM with an HMM to model the timbre spectra. These methods are described in much greater detail in section IV.
منابع مشابه
Embedding Songs and Tags for Playlist Prediction
Automatic playlist generation can be a useful tool to navigate the myriad choices available to users in music services today. Here, we present our recent work on explicitly modeling playlists without requiring external similarity measures. Our Logistic Markov Embedding is trained directly on historical playlist data and can unify songs and (when available) social tags in a Euclidean space. The ...
متن کاملDesign and Realisation of an Efficient Content Based Music Playlist Generation System
This thesis is on the subject of content based music playlist generation systems. The primary aim is to develop algorithms for content based music playlist generation that are faster than the current state of technology while keeping the quality of the playlists at a level that is at least comparable with that of the current state of technology. Not only need the algorithms be fast, they shall ...
متن کاملAutomated online playlist generation system
Received Sep 15 th , 2013 Revised Oct 25 th , 2013 Accepted Nov 30 th , 2013 This paper presents an online playlist management system designed to provide easy and intuitive management of audio and video playlists. Playlists can be uploaded, combined, downloaded, but also created online and stored for listening. Sharing of playlists is also enabled. State of the art AJAX technologies have been u...
متن کاملPAPA: Physiology and Purpose-Aware Automatic Playlist Generation
Abstract In this paper we present PAPA, a novel approach for automatically generating playlists. The proposed framework utilizes the user’s physiological response to music, together with traditional song meta-data to generate a playlist the user will not only enjoy, but which will assist him or her in achieving various user-defined goals (“purpose”). In addition to outlining the generic framewo...
متن کامل