logo资料库

Music Recommendation and Discovery.pdf

发布时间：2022-06-08 发布人：admin 分类：说明书 资料大小：7.37M 资料格式：pdf 举报 版权申诉
第1页 / 共252页
第2页 / 共252页
第3页 / 共252页
第4页 / 共252页
第5页 / 共252页
第6页 / 共252页
第7页 / 共252页
第8页 / 共252页
资料共252页，剩余部分请下载后查看
    Acknowledgements
    Abstract
    Resum
    Resumen
    Prologue
    Introduction
    Motivation
    Academia
    Industry
    The Problem
    The Solution
    Summary of contributions
    Thesis outline
    The recommendation problem
    Formalisation of the recommendation problem
    Use cases
    General model
    User profile representation
    Initial generation
    Maintenance
    Adaptation
    Recommendation methods
    Demographic filtering
    Collaborative filtering
    Content--based filtering
    Context--based filtering
    Hybrid methods
    Factors affecting the recommendation problem
    Summary
    Music recommendation
    Use Cases
    Artist recommendation
    Neighbour recommendation
    Playlist generation
    User profile representation
    Type of listeners
    Related work
    User profile representation proposals
    Item profile representation
    The music information plane
    Editorial metadata
    Cultural metadata
    Acoustic metadata
    Recommendation methods
    Collaborative filtering
    Content--based filtering
    Context--based filtering
    Hybrid methods
    Summary
    The Long Tail in recommender systems
    Introduction
    The Music Long Tail
    Definitions
    Qualitative, informal definition
    Quantitative, formal definition
    Qualitative versus quantitative definition
    Characterising a Long Tail distribution
    The dynamics of the Long Tail
    Novelty, familiarity and relevance
    Recommending the unknown
    Related work
    Summary
    Evaluation metrics
    Evaluation strategies
    System--centric evaluation
    Predictive--based metrics
    Decision--based metrics
    Rank--based metrics
    Other metrics
    Limitations
    Network--centric evaluation
    Navigation
    Connectivity
    Clustering
    Related work in music information retrieval
    Limitations
    User--centric evaluation
    Metrics
    Limitations
    Summary
    Network--centric evaluation
    Network analysis and the Long Tail model
    Artist network analysis
    Datasets
    Network analysis
    Popularity analysis
    Discussion
    User network analysis
    Datasets
    Network analysis
    Popularity analysis
    Discussion
    Summary
    User--centric evaluation
    Music Recommendation Survey
    Procedure
    Datasets
    Participants
    Results
    Participants
    Music Recommendation
    Discussion
    Limitations
    Applications
    Searchsounds: Music discovery in the Long Tail
    Motivation
    Goals
    System overview
    Summary
    FOAFing the Music: Music recommendation in the Long Tail
    Motivation
    Goals
    System overview
    Summary
    Conclusions and Further Research
    Summary of the Research
    Scientific contributions
    Industrial contributions
    Limitations and Further Research
    Outlook
    Appendix A. Publications
    Bibliography
    MUSIC RECOMMENDATION AND DISCOVERY IN THE LONG TAIL `Oscar Celma Herrada 2008
    c Copyright by `Oscar Celma Herrada 2008 All Rights Reserved ii
    who bring the whole endeavour into perspective. To Alex and Claudia iii
    iv
    Acknowledgements I would like to thank my supervisor, Dr. Xavier Serra, for giving me the opportunity to work on this very fascinating topic at the Music Technology Group (MTG). Also, I want to thank Perfecto Herrera for providing support, countless suggestions, reading all my writings, giving ideas, and devoting much time to me during this long journey. This thesis would not exist if it weren’t for the the help and assistance of many people. At the risk of unfair omission, I want to express my gratitude to them. I would like to thank all the colleagues from MTG that were —directly or indirectly— involved in some bits of this work. Special mention goes to Mohamed Sordo, Koppi, Pedro Cano, Mart´ın Blech, Emilia G´omez, Dmitry Bogdanov, Owen Meyers, Jens Grivolla, Cyril Laurier, Nicolas Wack, Xavier Oliver, Vegar Sandvold, Jos´e Pedro Garc´ıa, Nicolas Falquet, David Garc´ıa, Miquel Ram´ırez, and Otto W¨ust. Also, I thank the MTG/IUA Administration Staff (Cristina Garrido, Joana Clotet and Salvador Gurrera), and the sysadmins (Guillem Serrate, Jordi Funollet, Maarten de Boer, Ram´on Loureiro, and Carlos Atance). They provided help, hints and patience when I played around with the machines. During my six months stage at the Center for Computing Research of the National Poly- technic Institute (Mexico City) in 2007, I met a lot of interesting people ranging different disciplines. I thank Alexander Gelbukh for inviting me to work in his research group, the Natural Language Laboratory. Also, I would like to thank Grigori Sidorov, Tine Stalmans, Obdulia Pichardo, Sulema Torres, and Yulema Ledeneva for making my stay so wonderful. This thesis would be much more difficult to read —except for the “Spanglish” experts— if it weren’t for the excellent work of the following people: Paul Lamere, Owen Meyers, Terry Jones, Kurt Jacobson, Douglas Turnbull, Tom Slee, Kalevi Kilkki, Perfecto Herrera, Alberto Lumbreras, Daniel McEnnis, Xavier Amatriain, and Neil Lathia. They not only have helped me to improve the text, but have provided feedback, comments, suggestions, and —of course— criticism. v
    Many people have influenced my research during these years. Furthermore, I have been lucky enough to meet some of them. In this sense, I would like to acknowledge Elias Pampalk, Paul Lamere, Justin Donaldson, Jeremy Pickens, Markus Schedl, Peter Knees, and Stephan Baumann. I had very interesting discussions with them in several ISMIR (and other) conferences. Other researchers whom I have learnt a lot, and I have worked with, are: Massimiliano Zanin, Javier Buld´u, Rapha¨el Troncy, Michael Hausenblas, Roberto Garc´ıa, and Yves Raimond. I also want to thank some MTG veterans, whom I met and worked before starting the PhD. They are: Alex Loscos, Jordi Bonada, Pedro Cano, Oscar Mayor, Jordi Janer, Lars Fabig, Fabien Gouyon, and Enric Mieza. Also, special thanks goes to Esteban Maestre and Pau Arum´ı for having such a great time while being PhD students. Last but not least, this work would have never been possible without the encouragement of my wife Claudia, who has provided me love and patience, and my lovely son Alex —who altered my last.fm and youtube accounts with his favourite music. Nowadays, Cri–Cri, Elmo and Barney, coexists with The Dogs d’Amour, Backyard Babies, and other rock bands. I reckon that the two systems are a bit lost when trying to recommend me music and videos!. Also, a special warm thanks goes to my parents Tere and Toni, my brother Marc and my sister in law Marta, and the whole family in Barcelona and Mexico. At least, they will understand what my work is about. . . hopefully. This research was performed at the Music Technology Group of the Universitat Pompeu Fabra in Barcelona, Spain. Primary support was provided by the EU projects FP6-507142 SIMAC1 and FP6-045035 PHAROS2, and by a Mexican grant from the Secretar´ıa de Rela- ciones Exteriores (Ministry of Foreign Affairs) for a six months stage at the Center for Computing Research of the National Polytechnic Institute (Mexico City). 1http://www.semanticaudio.org 2http://www.pharos-audiovisual-search.eu/ vi
    Abstract Music consumption is biased towards a few popular artists. For instance, in 2007 only 1% of all digital tracks accounted for 80% of all sales. Similarly, 1,000 albums accounted for 50% of all album sales, and 80% of all albums sold were purchased less than 100 times. There is a need to assist people to filter, discover, personalise and recommend from the huge amount of music content available along the Long Tail. Current music recommendation algorithms try to accurately predict what people de- mand to listen to. However, quite often these algorithms tend to recommend popular —or well–known to the user— music, decreasing the effectiveness of the recommendations. These approaches focus on improving the accuracy of the recommendations. That is, try to make accurate predictions about what a user could listen to, or buy next, independently of how useful to the user could be the provided recommendations. In this Thesis we stress the importance of the user’s perceived quality of the recom- mendations. We model the Long Tail curve of artist popularity to predict —potentially— interesting and unknown music, hidden in the tail of the popularity curve. Effective recom- mendation systems should promote novel and relevant material (non–obvious recommenda- tions), taken primarily from the tail of a popularity distribution. The main contributions of this Thesis are: (i) a novel network–based approach for recommender systems, based on the analysis of the item (or user) similarity graph, and the popularity of the items, (ii) a user–centric evaluation that measures the user’s relevance and novelty of the recommendations, and (iii) two prototype systems that implement the ideas derived from the theoretical work. Our findings have significant implications for recommender systems that assist users to explore the Long Tail, digging for content they might like. vii
    viii
    分享到：
    赞 收藏

    相关推荐

    资料库

    课程资源

    共收录17145份资料，累计13个分类，关注成员有19位，主要包括：PHP,网络管理,网页制作,Java,.Net,数据库,3G/移动开发,C/C++,游戏开发,嵌入式,讲义,软件测试,专业指导

    热门标签

    PHP 网络管理 网页制作 Java .Net 数据库 3G/移动开发 C/C++ 游戏开发 嵌入式 讲义 软件测试 专业指导

    最新资料