September 26 th , 2010 Barcelona, Spain Copyright Information Copyright (c). This is an open-access article distributed under the terms of the Creative Commons Attribution License 3.0 Unported 1, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 http://creativecommons.org/licenses/by/3.0/ ii Organizing Committee Workshop Organizers • Amélie Anglade - Centre for Digital Music, Queen Mary, University of London, UK • Claudio Baccigalupo - Earbits, Los Angeles, CA, US • Norman Casagrande - last.fm, London, UK • Òscar Celma - BMAT, Barcelona, Spain • Paul Lamere - The Echo Nest, Somerville, MA, US Program Committee • Mathieu Barthet (Centre for Digital Music, Queen Mary, University of London) • Klaas Bosteels (last.fm) • Sally Jo Cunningham (Waikato University) • Justin Donaldson (Indiana University) • Benjamin Fields (Goldsmiths, University of London) • Emilia Gomez (Music Technology Group, Universitat Pompeu Fabra) • Fabien Gouyon (INESC Porto) • Peter Knees (Johannes Kepler University Linz) • Neal Lathia (University College London) • Daniel Lemire (Open Scholar, Montréal) • Yves Raimond (BBC Audio & Music interactive) • Markus Schedl (Johannes Kepler University Linz) • Mohamed Sordo (Music Technology Group, Universitat Pompeu Fabra) iii Preface Welcome to WOMRAD, the Workshop on Music Recommendation and Discovery being held in conjunction with ACM RecSys. In the last twenty years, there has been an amazing transformation in the world of music. Portable listening devices have advanced from the Sony Walkman that allowed you to carry ten songs in your pocket to the latest iPhone that can put millions of songs in your pocket via music subscription services such as Spotify or Rhapsody. Twenty years ago a typical personal music collection numbered around a thousand songs. Today, a music listener has access to millions of songs, drawn from all styles and genres from all over the world. The seemingly infinite choice today’s music listener faces can lead to a rich music listening experience, but only if the listener can find music that they want to listen to. Traditionally, music recommender systems have focused on the somewhat narrow task of attempting to predict a set of new artists or tracks for purchase or listening. Commerce sites like iTunes use music recommendation as a way to increase sales. Internet radio sites like Pandora use music recommendation as a way to offer personalized radio to millions of listeners. The success of music recommendation at iTunes and Pandora has led some to suggest that ‘music recommendation is solved’. Indeed, for narrow use cases like improving sales in a mainstream music store, or for creating satisfactory personalized radio streams, music recommendation may be good enough. However, this does not mean that music recommendation is solved. As music listeners spend more time interacting with multi-million song music collections, the need for tools that help listeners manage their listening will become increasingly important. Tools for exploring and discovering music especially in the long tail, tools for organizing listening, tools for creating interesting playlists, tools for managing group listening will all be essential to the music listening experience. Music recommendation technologies will be critical to building these tools. The WOMRAD workshop focuses on next generation of music recommender systems. Accepted papers fall into five categories: • Time Dependency - 1 paper - explorations in temporal patterns of music listening • Social Tagging - 3 papers - how semantic tags can be used to explain, compare and steer music recommender systems • Human-Computer Interaction - 2 papers - how music listeners interact with music and music recommender systems • Content-based Recommendation - 2 papers - techniques for recommendation base on audio content • Long Tail - 2 papers - how can systems make effective recommendations of new or unpopular content We are pleased to offer this selection of papers and hope that it serves as evidence that there is much interesting and fruitful research to be done in the area of music recommendation and discovery. We offer our thanks to all of the authors who submitted papers to this workshop. The Organizers, October 2010 iv Keynote Presentation The Dark Art: Is Music Recommendation Science a Science? Michael S. Papish , Product Development Director, Rovi Corporation Music preferences are emotional, subjective and full of social and cultural meaning. Practical experience building industrial recommendation applications suggests that user "trust" (a fuzzy concept combining user psychology with UI design and presentation) often overshadows actual results. What if making good music recommendations is actually a Dark Art and not a foundational problem of Information Retrieval Science? By tracing the beginnings of MIR, we present an early attempt at a Philosophy of Recommendation Science which tries to answer: • Does recommendation science exist only as a practical application? • Is it possible ground-truth metrics such as those proposed in the ISMIR 2001 Resolution don't actually exist? • What types of solvable scientific problems should receive academic attention from the MIR community? • Cee Lo's Teeth: Scariest in the entire history of recorded music? v Table of Contents Time Dependency Rocking around the clock eight days a week: an exploration of temporal patterns of music listening Perfecto Herrera, Zuriñe Resa and Mohamed Sordo ....................................................... 1 Social Tagging Using Song Social Tags and Topic Models to Describe and Compare Playlists Benjamin Fields, Christophe Rhodes and Mark d'Inverno ............................................... 5 Piloted Search and Recommendation with Social Tag Cloud-Based Navigation Cédric Mesnage and Mark Carman ................................................................................ 13 A Method for Obtaining Semantic Facets of Music Tags Mohamed Sordo, Fabien Gouyon and Luís Sarmento .................................................... 21 Computer-Human Interaction A Survey of Recommendation Aids Pirkka Åman and Lassi Liikkanen ................................................................................... 25 The Role People Play in Adolescents’ Music Information Acquisition Audrey Laplante .............................................................................................................. 29 Content-based Recommendation Content-based music recommendation based on user preference examples Dmitry Bogdanov, Martín Haro, Ferdinand Fuhrmann, Emilia Gómez and Perfecto Herrera ............................................................................................................. 33 Applying Constrained Clustering for Active Exploration of Music Collections Pedro Mercado and Hanna Lukashevich ........................................................................ 39 Long Tail Music Recommendation in the Personal Long Tail: Using a Social-based Analysis of a User’s Long-Tailed Listening Behavior Kibeom Lee, Woon Seung Yeo and Kyogu Lee ............................................................... 47 Music Recommendation and the Long Tail Mark Levy and Klaas Bosteels ........................................................................................ 55 vi Rocking around the clock eight days a week: an exploration of temporal patterns of music listening Perfecto Herrera Zuriñe Resa Mohamed Sordo Music Technology Group Department of Technology Universitat Pompeu Fabra [email protected] [email protected] [email protected] ABSTRACT systems. Music listening decisions might seem expressions of free Music listening patterns can be influenced by contextual factors will but they are in fact influenced by interlinked social, such as the activity a listener is involved in, the place one is environmental, cognitive and biological factors [21][22]. located or physiological constants. As a consequence, musical Chronobiology is the discipline that deals with time and rhythm in listening choices might show some recurrent temporal patterns. living organisms. The influence of circadian rhythms (those Here we address the hypothesis that for some listeners, the showing a repetition pattern every 24 hours approximately, selection of artists and genres could show a preference for certain usually linked to the day-night alternation), but also of ultradian moments of the day or for certain days of the week. With the help rhythms (those recurring in a temporal lag larger than one day like of circular statistics we analyze playcounts from Last.fm and the alternation of work and leisure or the seasons), has been detect the existence of that kind of patterns. Once temporal demonstrated on different levels of organization of many living preference is modeled for each listener, we test the robustness of creatures, and preserving some biological cycles is critical to keep that using the listener’s playcount from a posterior temporal an optimum health [18]. The observation that human behavior is period. We show that for certain users, artists and genres, modulated by rhythms of hormonal releases, exposure to light, temporal patterns of listening can be used to predict music weather conditions, moods, and also by the activity we are listening selections with above-chance accuracy. This finding engaged into [12][3] paves the way to our main hypothesis: there could be exploited in music recommendation and playlist