Recommender Systemsfor Social Tagging SystemsSocial tagging systems are Web 2.0 applications that promote user participation through facilitated content sharing and annotation of that content with freely chosen keywords, called tags. Despite the potential of social tagging to improve organization an
Leandro balby marinho. Andreas hotho Robert jaschke Alexandros nanopoulos Steffen rendle lars schmidt-Thieme Gerd Stumme. Panagiotis Symeonidis Recommender systems for Social Tagging Systems ② Springe er Leandro balby marinho Andreas hotho Federal University of Campina grande University of Wurzburg Brazil Ibmarinho@dsc. ufcg edu br Robert jaschke Alexandros Nanopoulos University of Kassel University of Hildesheim Germany Germany Steffen rendle Lars schmidt -Thieme University of Konstanz University of hildesheim Germany Germany Gerd Stumme Panagiotis Symeonidis University asse Aristotle University ermany greece ISSN2191-8112 e-ISsN2191-8120 ISBN978-1-4614-1893-1 e-ISBN978-1-4614-1894-8 DOI10.1007/978-1-4614-1894-8 Springer New York Dordrecht Heidelberg london Library of Congress Control Number: 2012931154 C The Author(s)2012 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher(Springer Science+ Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights Printed on acid-free paper SpringerispartofSpringerScience+BusinessMedia(www.springer.com) Preface Social tagging systems are Web 2.0 applications that promote user participa tion through facilitated content sharing and annotation of that content with freely chosen keywords, called tags. Despite the potential of social tagging to improve organization and sharing of content, without efficient tools for con- tent filtering and search, users are prone to suffer from information overload as more and more users, content, and tags become available on-line. Recom mender systems are among the best known techniques for helping users to filter out and discover relevant information in large datasets. However, social tagging systems put forward new challenges for recommender systems since differently from the standard recommender setting where users are mainly interested in content- in social tagging systems users may additionally be interested in finding tags and even other users The goal of this book is to bring together important research in a new fam lly of recommender systems aimed at serving social tagging systems. While y no means exhaustive, the chapters introduce a wide variety of recent ap- proaches, from the most basic to the state-of-the-art, for providing recom mendations in social tagging systems. The focus is on tag recommendations and tag-aware recommendations, which are the prevalent recommendation tasks in the literature and real-world social tagging systems. The material covered in the book is aimed at graduate students, teachers, researchers, and practitioners in the areas of web mining, e-commerce, information retrieval and machine learning The idea for this book emerged from a long history of fruitful cooperation between the authors, who have been actively contributing in many of the topics covered in this book. Many parts of the book are built on top of the authors' previous book chapter entitled Social Tagging Recommender Systems published in the Recommender Systems Handbook in 2011; which triggered the cooperation with Springer for extending it into a book The book is organized into three parts. Part I provides introductory ma- terial on social tagging systems and recommender systems. Part II presents a wide variety of recommendation techniques, ranging from the most basic Preface methods to the state-of-the-art, as well as strategies for evaluating these rec ommender systems. Part IiI provides a detailed case study on the technical aspects of deploying and evaluating recommender systems in BibSonomy, a real-world social tagging system of bookmarks and scientific references Contents Part I foundations 1 Social Tagging Systems 1.1 Introduction 1.2 Folksonomies 1.3 Tag Clouds 1. 4 Data Representation 1.4.1 Folksonomies as Tensors 1.4.2 Folksonomies as Hypergraphs 334556678 1.5 Recommendation Tasks in STS 1.5.1 User Recommendation 1.5.2 Resource recommendation 1.5.3 Tag Recommendation 1.6 Recommendations in Social Tagging Systems ...10 1.6.1 Bibsonomy 10 1.6.2 CiteULike 11 1.6.3 Other Systems 12 1. 7 Notation 1. 8 Further reading 14 References 14 2 Recommender Systems 17 2.1 Rating and Item Prediction 17 2.2 Rating Prediction as Regression Problem 18 2.3 Item Prediction as Ranking problem .21 2.4 User and Item Attributes 23 2.5 New User and New Item Problems 24 2.6 Context-aware and Multi-Mode recommendations References 27 11 Contents Part II Recommendation Techniques for Social Tagging Systems 3 Baseline Techniques 33 3.1 Constant Models 33 3.1.1 Tag Recommendation 33 3.1.2 User/Tag-aware Recommendation 34 3.1.3 Remarks on Complexity 34 3.2 Projection Matrices 3.3 Projection-based Collaborative Filtering 35 36 3.3.1 Tag Recommendations 36 3.3.2 Tag-aware Recommendations 37 3.3.3 User recommendations 38 3.3.4 Remarks on Complexity 39 3. 4 Further reading 39 References 41 4 Advanced Techniques .43 4.1 Factorization models 43 4.1.1 Higher Order Singular Value Decomposition HOSVD on Tensors 4.1.2 Scalable Factorization models 51 4.1.3 Learning Tag Recommendation Models ) 4.2 Graph-based Models 57 4.2.1 PageRank-based Recommendations in STS 4.2.2 Relational Neighbors for Tag Recommendations 4.3 Content and social-Based models 63 4.3.1 Exploiting the Content of resources 4.3.2 Exploiting Social Relations 65 4.4 Further readin 69 References 70 5 ffline evaluation 5.1 Evaluation metrics 5.1.1 Precision and recall 5.1.2 Further measures 76 5.2 Evaluation Protocols 76 5.2.1 LeavePostOut Methodology 76 5.2.2 Time-based Splits 5.3 Comparison of Tag Re recommenders 77 References Part III Implementing Recommender Systems for Social Tagging 6 Real World Social Tagging Recommender systems 6.1 Introduction 6.2 Challenges and requirements 85 Contents 6.3 The BibSonomy Social Tagging System 86 6.4 Architecture 87 6.4.1 Overview 87 6.4.2 Recommender interface 88 6.4.3 Logging 6.5 Recommender Implementations 6.5.1 Meta Recommender 90 6.5.2 Multiplexing Tag recommender 6.5.3 Example recommender Implementations 92 6.6 Further reading References 94 7 Online evaluation 97 7.1 Evaluation Setting 97 7.1.1 Metrics and Protocols 97 7.1.2 Preprocessing and Cleansing 97 7. 2 Case Study ..98 7.2.1 General Results 98 7.2.2 Influence of the reload, Button 7.2.3 Logged click' Events 7.2.4 Average F1-Measure per User 102 7.3 The ECML PKDD Discovery Challenge 2009 103 7.3.1 Setting 104 7.3.2 Methods ..104 7.3.3 Results .105 7.4 Conclusion 107 References 108 8 Conclusions 8.1 Summary.… 109 8.2 Discussion and Outlook 110 References 111