Music Recommendation and Discovery : The Long Tail, Long Fail, and Long Play in the Digital Music Space

In the last 15 years we have seen a major transformation in the world of music. - sicians use inexpensive personal computers instead of expensive recording studios to record, mix and engineer music. Musicians use the Internet to distribute their - sic for free instead of spending large amounts of money creating CDs, hiring trucks and shipping them to hundreds of record stores. As the cost to create and distribute recorded music has dropped, the amount of available music has grown dramatically. Twenty years ago a typical record store would have music by less than ten thousand artists, while today online music stores have music catalogs by nearly a million artists. While the amount of new music has grown, some of the traditional ways of ?nding music have diminished. Thirty years ago, the local radio DJ was a music tastemaker, ?nding new and interesting music for the local radio audience. Now - dio shows are programmed by large corporations that create playlists drawn from a limited pool of tracks. Similarly, record stores have been replaced by big box reta- ers that have ever-shrinking music departments. In the past, you could always ask the owner of the record store for music recommendations. You would learn what was new, what was good and what was selling. Now, however, you can no longer expect that the teenager behind the cash register will be an expert in new music, or even be someone who listens to music at all. Foreword Preface Acknowledgements Contents 1 Introduction 1.1 Motivation 1.1.1 Academia 1.1.2 Industry 1.2 What's the Problem with Music Recommendation? 1.2.1 Music = Movies and Books 1.2.2 Predictive Accuracy vs. Perceived Quality 1.3 Our Proposal 1.3.1 Novelty and Relevance 1.3.2 Key Elements 1.4 Summary of Contributions 1.5 Book Outline References 2 The Recommendation Problem 2.1 Formalisation of the Recommendation Problem 2.2 Use Cases 2.3 General Model 2.4 User Profile Representation 2.4.1 Initial Generation 2.4.2 Maintenance 2.4.3 Adaptation 2.5 Recommendation Methods 2.5.1 Demographic Filtering 2.5.2 Collaborative Filtering 2.5.3 Content-Based Filtering 2.5.4 Context-Based Filtering 2.5.5 Hybrid Methods 2.6 Factors Affecting the Recommendation Problem 2.6.1 Novelty and Serendipity 2.6.2 Explainability 2.6.3 Cold Start Problem 2.6.4 Data Sparsity and High Dimensionality 2.6.5 Coverage 2.6.6 Trust 2.6.7 Attacks 2.6.8 Temporal Effects 2.6.9 Understanding the Users 2.7 Summary References 3 Music Recommendation 3.1 Use Cases 3.1.1 Artist Recommendation 3.1.2 Playlist Generation 3.1.3 Neighbour Recommendation 3.2 User Profile Representation 3.2.1 Type of Listeners 3.2.2 Related Work 3.2.3 User Profile Representation Proposals 3.3 Item Profile Representation 3.3.1 The Music Information Plane 3.3.2 Editorial Metadata 3.3.3 Cultural Metadata 3.3.4 Acoustic Metadata 3.4 Recommendation Methods 3.4.1 Collaborative Filtering 3.4.2 Context-Based Filtering 3.4.3 Content-Based Filtering 3.4.4 Hybrid Methods 3.5 Summary 3.5.1 Links with the Following Chapters References 4 The Long Tail in Recommender Systems 4.1 Introduction 4.1.1 Pre- and post-filters 4.2 The Music Long Tail 4.2.1 The Long Tail of Sales Versus the Long Tail of Plays 4.2.2 Collecting Playcounts for the Music Long Tail 4.2.3 An Example 4.3 Definitions 4.3.1 Qualitative, Informal Definition 4.3.2 Quantitative, Formal Definition 4.3.3 Qualitative Versus Quantitative Definition 4.4 Characterising a Long Tail Distribution 4.4.1 Not All Long Tails Are Power-Law 4.4.2 A Model Selection: Power-Law or Not Power-Law? 4.5 The Dynamics of the Long Tail 4.5.1 Strike a Chord? 4.6 Novelty, Familiarity and Relevance 4.6.1 Recommending the Unknown 4.6.2 Related Work 4.7 Summary 4.7.1 Links with the Following Chapters References 5 Evaluation Metrics 5.1 Evaluation Strategies 5.2 System-Centric Evaluation 5.2.1 Predictive-Based Metrics 5.2.2 Decision-Based Metrics 5.2.3 Rank-Based Metrics 5.2.4 Limitations 5.3 Network-Centric Evaluation 5.3.1 Complex Network Analysis 5.3.2 Navigation 5.3.3 Connectivity 5.3.4 Clustering 5.3.5 Centrality 5.3.6 Limitations 5.3.7 Related Work in Music Information Retrieval 5.4 User-Centric Evaluation 5.4.1 Gathering Feedback 5.4.2 Limitations 5.5 Summary 5.5.1 Links with the Following Chapters References 6 Network-Centric Evaluation 6.1 Network Analysis and the Long Tail Model 6.2 Artist Network Analysis 6.2.1 Datasets 6.2.2 Network Analysis 6.2.3 Popularity Analysis 6.2.4 Discussion 6.3 User Network Analysis 6.3.1 Datasets 6.3.2 Network Analysis 6.3.3 Popularity Analysis 6.3.4 Discussion 6.4 Summary 6.4.1 Links with the Following Chapters References 7 User-Centric Evaluation 7.1 Music Recommendation Survey 7.1.1 Procedure 7.1.2 Datasets 7.1.3 Participants 7.2 Results 7.2.1 Demographic Data 7.2.2 Quality of the Recommendations 7.3 Discussion 7.4 Limitations 8 Applications 8.1 Searchsounds: Music Discovery in the Long Tail 8.1.1 Motivation 8.1.2 Goals 8.1.3 System Overview 8.1.4 Summary 8.2 FOAFing the Music: Music Recommendation in the Long Tail 8.2.1 Motivation 8.2.2 Goals 8.2.3 System Overview 8.2.4 Summary References 9 Conclusions and Further Research 9.1 Book Summary 9.1.1 Scientific Contributions 9.1.2 Industrial Contributions 9.2 Limitations and Further Research 9.2.1 Dynamic Versus Static Data 9.2.2 Domain Specific 9.2.3 User Evaluation 9.2.4 User Understanding 9.2.5 Recommendations with No Explanation 9.3 Outlook References Index