IT training LNCS 9077 advances in knowledge discovery and data mining (part 1) cao, lim, zhou, ho, cheung motoda 2015 04 14

Honorary Co-chairsPhan Thanh Binh Vietnam National University, Ho Chi Minh City, VietnamMasaru Kitsuregawa National Institute of Informatics, Japan General Co-chairs Tu-Bao Ho Japan Adva

Tru Cao · Ee-Peng Lim

Zhi-Hua Zhou · Tu-Bao Ho

123

19th Pacific-Asia Conference, PAKDD 2015

Ho Chi Minh City, Vietnam, May 19–22, 2015

Proceedings, Part I

Advances in

Knowledge Discovery and Data Mining

Subseries of Lecture Notes in Computer Science

LNAI Series Editors

DFKI and Saarland University, Saarbrücken, Germany

LNAI Founding Series Editor

Joerg Siekmann

DFKI and Saarland University, Saarbrücken, Germany

Tru Cao · Ee-Peng Lim

Zhi-Hua Zhou · Tu-Bao Ho

David Cheung · Hiroshi Motoda (Eds.)

Advances in

Knowledge Discovery

and Data Mining

19th Pacific-Asia Conference, PAKDD 2015

Ho Chi Minh City, Vietnam, May 19–22, 2015 Proceedings, Part I

ABC

Tru Cao

Ho Chi Minh City University of Technology

Ho Chi Minh City

Nomi CityJapanDavid CheungThe University of Hong KongHong Kong

Hong Kong SARHiroshi MotodaOsaka UniversityOsaka

Japan

Lecture Notes in Artificial Intelligence

DOI 10.1007/978-3-319-18038-0

Library of Congress Control Number: 2015936624

LNCS Sublibrary: SL7 – Artificial Intelligence

Springer Cham Heidelberg New York Dordrecht London

c

 Springer International Publishing Switzerland 2015

This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad- casting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known

or hereafter developed.

The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)

After ten years since PAKDD 2005 in Ha Noi, PAKDD was held again in Vietnam,during May 19–22, 2015, in Ho Chi Minh City PAKDD 2015 is the 19th edition of thePacific-Asia Conference series on Knowledge Discovery and Data Mining, a leadinginternational conference in the field The conference provides a forum for researchersand practitioners to present and discuss new research results and practical applications.There were 405 papers submitted to PAKDD 2015 and they underwent a rigorousdouble-blind review process Each paper was reviewed by three Program Committee(PC) members in the first round and meta-reviewed by one Senior Program Committee(SPC) member who also conducted discussions with the reviewers The Program Chairsthen considered the recommendations from SPC members, looked into each paper andits reviews, to make final paper selections At the end, 117 papers were selected for theconference program and proceedings, resulting in the acceptance rate of 28.9%, amongwhich 26 papers were given long presentation and 91 papers given regular presentation.The conference started with a day of six high-quality workshops During the nextthree days, the Technical Program included 20 paper presentation sessions coveringvarious subjects of knowledge discovery and data mining, three tutorials, a data min-ing contest, a panel discussion, and especially three keynote talks by world-renownedexperts.

PAKDD 2015 would not have been so successful without the efforts, contributions,and supports by many individuals and organizations We sincerely thank the HonoraryChairs, Phan Thanh Binh and Masaru Kitsuregawa, for their kind advice and supportduring preparation of the conference We would also like to thank Masashi Sugiyama,Xuan-Long Nguyen, and Thorsten Joachims for giving interesting and inspiring keynotetalks

We would like to thank all the Program Committee members and external reviewersfor their hard work to provide timely and comprehensive reviews and recommenda-tions, which were crucial to the final paper selection and production of the high-qualityTechnical Program We would also like to express our sincere thanks to the followingOrganizing Committee members: Xiaoli Li and Myra Spiliopoulou together with the in-dividual Workshop Chairs for organizing the workshops; Dinh Phung and U Kang withthe tutorial speakers for arranging the tutorials; Hung Son Nguyen, Nitesh Chawla, andNguyen Duc Dung for running the contest; Takashi Washio and Jaideep Srivastava forpublicizing to attract submissions and participants to the conference; Tran Minh-Trietand Vo Thi Ngoc Chau for handling the whole registration process; Tuyen N Huynh forcompiling all the accepted papers and for working with the Springer team to producethese proceedings; and Bich-Thuy T Dong, Bac Le, Thanh-Tho Quan, and Do Phuc forthe local arrangements to make the conference go smoothly

We are grateful to all the sponsors of the conference, in particular AFOSR/AOARD(Air Force Office of Scientific Research/Asian Office of Aerospace Research and Devel-opment), for their generous sponsorship and support, and the PAKDD Steering

Committee for its guidance and Student Travel Award and Early Career Research Awardsponsorship We would also like to express our gratitude to John von Neumann Insti-tute, University of Technology, University of Science, and University of InformationTechnology of Vietnam National University at Ho Chi Minh City and Japan AdvancedInstitute of Science and Technology for jointly hosting and organizing this conference.Last but not least, our sincere thanks go to all the local team members and volunteeringhelpers for their hard work to make the event possible.

We hope you have enjoyed PAKDD 2015 and your time in Ho Chi Minh City,Vietnam

Ee-Peng LimZhi-Hua ZhouTu-Bao HoDavid CheungHiroshi Motoda

Honorary Co-chairs

Phan Thanh Binh Vietnam National University, Ho Chi Minh City,

VietnamMasaru Kitsuregawa National Institute of Informatics, Japan

General Co-chairs

Tu-Bao Ho Japan Advanced Institute of Science and

Technology, JapanDavid Cheung University of Hong Kong, China

Hiroshi Motoda Institute of Scientific and Industrial Research,

Osaka University, Japan

Program Committee Co-chairs

Tru Hoang Cao Ho Chi Minh City University of Technology,

VietnamEe-Peng Lim Singapore Management University, SingaporeZhi-Hua Zhou Nanjing University, China

Tutorial Co-chairs

Dinh Phung Deakin University, Australia

U Kang Korea Advanced Institute of Science and

Technology, Korea

Workshop Co-chairs

Xiaoli Li Institute for Infocomm Research, A*STAR,

SingaporeMyra Spiliopoulou Otto-von-Guericke University Magdeburg,

Germany

Publicity Co-chairs

Takashi Washio Institute of Scientific and Industrial Research,

Osaka University, JapanJaideep Srivastava University of Minnesota, USA

Proceedings Chair

Tuyen N Huynh John von Neumann Institute, Vietnam

Contest Co-chairs

Hung Son Nguyen University of Warsaw, Poland

Nitesh Chawla University of Notre Dame, USA

Nguyen Duc Dung Vietnam Academy of Science and Technology,

Vietnam

Local Arrangement Co-chairs

Bich-Thuy T Dong John von Neumann Institute, Vietnam

Bac Le Ho Chi Minh City University of Science, VietnamThanh-Tho Quan Ho Chi Minh City University of Technology,

Vietnam

Do Phuc University of Information Technology, Vietnam

National University at Ho Chi Minh City,Vietnam

Treasurer

Graham Williams Togaware, Australia

Tu-Bao Ho Japan Advanced Institute of Science and

Technology, Japan (Member since 2005,Co-chair 2012–2014, Chair 2015–2017,Life Member since 2013)

Ee-Peng Lim (Co-chair) Singapore Management University, Singapore

(Member since 2006, Co-chair 2015–2017)Jaideep Srivastava University of Minnesota, USA (Member

since 2006)Zhi-Hua Zhou Nanjing University, China (Member since 2007)Takashi Washio Institute of Scientific and Industrial Research,

Osaka University, Japan (Member since 2008)Thanaruk Theeramunkong Thammasat University, Thailand (Member

since 2009)

P Krishna Reddy International Institute of Information Technology,

Hyderabad (IIIT-H), India (Member since 2010)Joshua Z Huang Shenzhen Institutes of Advanced Technology,

Chinese Academy of Sciences, China(Member since 2011)

Longbing Cao Advanced Analytics Institute, University of

Technology, Sydney, Australia(Member since 2013)

Jian Pei School of Computing Science, Simon Fraser

University, Canada (Member since 2013)Myra Spiliopoulou Otto-von-Guericke-University Magdeburg,

Germany (Member since 2013)Vincent S Tseng National Cheng Kung University, Taiwan

(Member since 2014)

Life Members

Hiroshi Motoda AFOSR/AOARD and Institute of Scientific and

Industrial Research, Osaka University, Japan(Member since 1997, Co-chair 2001–2003,Chair 2004–2006, Life Member since 2006)Rao Kotagiri University of Melbourne, Australia

(Member since 1997, Co-chair 2006–2008,Chair 2009–2011, Life Member since 2007)Huan Liu Arizona State University, USA (Member

since 1998, Treasurer 1998–2000, Life Membersince 2012)

Ning Zhong Maebashi Institute of Technology, Japan

(Member since 1999, Life member since 2008)Masaru Kitsuregawa Tokyo University, Japan (Member since 2000,

Life Member since 2008)David Cheung University of Hong Kong, China (Member since

2001, Treasurer 2005–2006,chair 2006–2008, Life Member since 2009)Graham Williams Australian National University, Australia

(Member since 2001, Treasurer since 2006,Co-chair 2009–2011, Chair 2012–2014,Life Member since 2009)

Ming-Syan Chen National Taiwan University, Taiwan, ROC

(Member since 2002, Life Member since 2010)Kyu-Young Whang Korea Advanced Institute of Science and

Technology, Korea (Member since 2003,Life Member since 2011)

Chengqi Zhang University of Technology, Sydney, Australia

(Member since 2004, Life Member since 2012)

Senior Program Committee Members

Arbee Chen National Chengchi University, Taiwan

Bart Goethals University of Antwerp, Belgium

Charles Ling University of Western Ontario, Canada

Chih-Jen Lin National Taiwan University, Taiwan

Dacheng Tao University of Technology, Sydney, AustraliaDou Shen Baidu, China

George Karypis University of Minnesota, USA

Haixun Wang Google, USA

Hanghang Tong City University of New York, USA

Hui Xiong Rutgers Univesity, USA

Ian Davidson University of California Davis, USA

James Bailey University of Melbourne, Australia

Jeffrey Yu The Chinese University of Hong Kong, Hong KongJian Pei Simon Fraser University, Canada

Jianyong Wang Tsinghua University, China

Jieping Ye Arizona State University, USA

Jiuyong Li University of South Australia, Australia

Joshua Huang Shenzhen Institutes of Advanced Technology,

Chinese Academy of Sciences, ChinaKyuseok Shim Seoul National University, Korea

Longbing Cao University of Technology, Sydney, AustraliaMasashi Sugiyama University of Tokyo, Japan

Michael Berthold University of Konstanz, Germany

Ming Li Nanjing University, China

Ming-Syan Chen National Taiwan University, Taiwan

Min-Ling Zhang Southeast University, China

Myra Spiliopoulou Otto-von-Guericke-University Magdeburg,

GermanyNikos Mamoulis University of Hong Kong, Hong Kong

Ning Zhong Maebashi Institute of Technology, Japan

Osmar Zaiane University of Alberta, Canada

P Krishna Reddy International Institute of Information Technology,

Hyderabad, IndiaPeter Christen Australian National University, Australia

Sanjay Chawla University of Sydney, Australia

Takashi Washio Institute of Scientific and Industrial Research,

Osaka University, JapanVincent S Tseng National Cheng Kung University, Taiwan

Wee Keong Ng Nanyang Technological University, SingaporeWei Wang University of California at Los Angeles, USAWen-Chih Peng National Chiao Tung University, Taiwan

Xiaofang Zhou University of Queensland, Australia

Xiaohua Hu Drexel University, USA

Xifeng Yan University of California, Santa Barbara, USAXindong Wu University of Vermont, USA

Xing Xie Microsoft Research Asia, China

Yanchun Zhang Victoria University, Australia

Yu Zheng Microsoft Research Asia, China

Program Committee Members

Aijun An York University, Canada

Aixin Sun Nanyang Technological University, SingaporeAkihiro Inokuchi Kwansei Gakuin University, Japan

Alfredo Cuzzocrea ICAR-CNR and University of Calabria, ItalyAndrzej Skowron University of Warsaw, Poland

Anne Denton North Dakota State University, USA

Bettina Berendt Katholieke Universiteit Leuven, Belgium

Bin Zhou University of Maryland, Baltimore County, USABing Tian Dai Singapore Management University, Singapore

Bo Zhang Tsinghua University, China

Bolin Ding Microsoft Research, USA

Bruno Cremilleux Université de Caen Basse-Normandie, FranceCarson K Leung University of Manitoba, Canada

Chandan Reddy Wayne State University, USA

Chedy Raissi Inria, France

Chengkai Li The University of Texas at Arlington, USA

Chia-Hui Chang National Central University, Taiwan

Chiranjib Bhattacharyya Indian Institute of Science, India

Choochart Haruechaiy National Electronics and Computer Technology

Center, ThailandChun-Hao Chen Tamkang University, Taiwan

Chun-hung Li Hong Kong Baptist University, Hong KongClifton Phua NCS, Singapore

Daoqiang Zhang Nanjing University of Aeronautics and

Astronautics, ChinaDao-Qing Dai Sun Yat-Sen University, China

David Taniar Monash University, Australia

David Lo Singapore Management University, SingaporeDe-Chuan Zhan Nanjing University, China

Dejing Dou University of Oregon, USA

De-Nian Yang Academia Sinica, Taiwan

Dhaval Patel Indian Institute of Technology, Roorkee, IndiaDinh Phung Deakin University, Australia

Dragan Gamberger Ru ¯der Boškovi´c Institute, Croatia

Du Zhang California State University, Sacramento, USADuc Dung Nguyen Institute of Information Technology, VietnamEnhong Chen University of Science and Technology of China,

ChinaFei Liu Carnegie Mellon University, USA

Feida Zhu Singapore Management University, SingaporeFlorent Masseglia Inria, France

Geng Li Oracle Corporation, USA

Giuseppe Manco Università della Calabria, Italy

Guandong Xu University of Technology, Sydney, AustraliaGuo-Cheng Lan Industrial Technology Research Institute, TaiwanGustavo Batista University of São Paulo, Brazil

Hady Lauw Singapore Management University, SingaporeHarry Zhang University of New Brunswick, Canada

Hiroshi Mamitsuka Kyoto University, Japan

Hong Shen University of Adelaide, Australia

Hsuan-Tien Lin National Taiwan University, Taiwan

Hua Lu Aalborg University, Denmark

Hui Wang University of Ulster, UK

Hung Son Nguyen University of Warsaw, Poland

Hung-Yu Kao National Cheng Kung University, Taiwan

Irena Koprinska University of Sydney, Australia

J Saketha Nath Indian Insitiute of Technology, India

Jaakko Hollmén Aalto University, Finland

Jake Chen Indiana University–Purdue University Indianapolis,

USA

James Kwok Hong Kong University of Science and Technology,

ChinaJason Wang New Jersey Science and Technology University,

USAJean-Marc Petit Université de Lyon, France

Jeffrey Ullman Stanford University, USA

Jen-Wei Huang National Cheng Kung University, Taiwan

Jerry Chun-Wei Lin Harbin Institute of Technology Shenzhen,

ChinaJia Wu University of Technology, Sydney, AustraliaJialie Shen Singapore Management University, SingaporeJiayu Zhou Samsung Research America, USA

Jia-Yu Pan Google, USA

Jin Soung Yoo Indiana University–Purdue University

Indianapolis, USAJingrui He IBM Research, USA

Jinyan Li University of Technology, Sydney, AustraliaJohn Keane University of Manchester, UK

Jun Huan University of Kansas, USA

Jun Gao Peking University, China

Jun Luo Huawei Noah’s Ark Lab, Hong Kong

Jun Zhu Tsinghua University, China

Junbin Gao Charles Sturt University, Australia

Junjie Wu Beihang University, China

Junping Zhang Fudan University, China

K Selcuk Candan Arizona State University, USA

Keith Chan Hong Kong Polytechnic University, Hong KongKhoat Than Hanoi University of Science and Technology,

VietnamKitsana Waiyamai Kasetsart University, Thailand

Krisztian Buza Semmelweis University, Budapest, HungaryKun-Ta Chuang National Cheng Kung University, Taiwan

Kuo-Wei Hsu National Chengchi University, Taiwan

Latifur Khan University of Texas at Dallas, USA

Ling Chen University of Technology, Sydney, AustraliaLipo Wang Nanyang Technological University, SingaporeManabu Okumura Japan Advanced Institute of Science and

Technology, JapanMarco Maggini Università degli Studi di Siena, Italy

Marian Vajtersic University of Salzburg, Austria

Marut Buranarach National Electronics and Computer Technology

Center, ThailandMary Elaine Califf Illinois State University, USA

Marzena Kryszkiewicz Warsaw University of Technology, Poland

Masashi Shimbo Nara Institute of Science and Technology, JapanMeng Chang Chen Academia Sinica, Taiwan

Mengjie Zhang Victoria University of Wellington, New ZealandMichael Hahsler Southern Methodist University, USA

Min Yao Zhejiang University, China

Mi-Yen Yeh Academia Sinica, Taiwan

Muhammad Cheema Monash University Australia

Murat Kantarcioglu University of Texas at Dallas, USA

Ngoc-Thanh Nguyen Wrocław University of Technology, PolandNguyen Le Minh Japan Advanced Institute of Science and

Technology, JapanPabitra Mitra Indian Institute of Technology Kharagpur, IndiaPatricia Riddle University of Auckland, New Zealand

Peixiang Zhao Florida State University, USA

Philippe Lenca Télécom Bretagne, France

Philippe Fournier-Viger University of Moncton, Canada

Qingshan Liu NLPR Institute of Automation, Chinese Academy

of Sciences, ChinaRaymond Chi-Wing Wong Hong Kong University of Science and Technology,

Hong KongRichi Nayak Queensland University of Technology, AustraliaRui Camacho Universidade do Porto, Portugal

Salvatore Orlando University of Venice, Italy

Sanjay Jain National University of Singapore, SingaporeSee-Kiong Ng Institute for Infocomm Research, A*STAR,

SingaporeShafiq Alam University of Auckland, New Zealand

Sheng-Jun Huang Nanjing University of Aeronautics and

Astronautics, ChinaShoji Hirano Shimane University, Japan

Shou-De Lin National Taiwan University, Taiwan

Shuai Ma Beihang University, China

Shu-Ching Chen Florida International University, USA

Shuigeng Zhou Fudan University, China

Silvia Chiusano Politecnico di Torino, Italy

Songcan Chen Nanjing University of Aeronautics and

Astronautics, ChinaTadashi Nomoto National Institute of Japanese Literature, JapanTakehisa Yairi University of Tokyo, Japan

Tetsuya Yoshida Nara Women’s University, Japan

Toshihiro Kamishima National Institute of Advanced Industrial Science

and Technology, Japan

Tuyen N Huynh John von Neumann Institute, Vietnam

Tzung-Pei Hong National University of Kaohsiung, Taiwan

Van-Nam Huynh Japan Advanced Institute of Science and

Technology, JapanVincenzo Piuri Università degli Studi di Milano, Italy

Wai Lam The Chinese University of Hong Kong, Hong KongWalter Kosters Universiteit Leiden, The Netherlands

Wang-Chien Lee Pennsylvania State University, USA

Wei Ding University of Massachusetts Boston, USA

Wenjie Zhang University of New South Wales, Australia

Wenjun Zhou University of Tennessee, Knoxville, USA

Wilfred Ng Hong Kong University of Science and Technology,

Hong KongWu-Jun Li Nanjing University, China

Wynne Hsu National University of Singapore, SingaporeXiaofeng Meng Renmin University of China, China

Xiaohui (Daniel) Tao University of Southern Queensland, AustraliaXiaoli Li Institute for Infocomm Research, A*STAR,

SingaporeXiaowei Ying Bank of America, USA

Xin Wang University of Calgary, Canada

Xingquan Zhu Florida Atlantic University, USA

Xintao Wu University of Arkansas, Arkansas

Xuan Vinh Nguyen University of Melbourne, Australia

Xuan-Hieu Phan University of Engineering and

Technology–Vietnam National University,Hanoi, Vietnam

Xuelong Li University of London, UK

Xu-Ying Liu Southeast University, China

Yang Yu Nanjing University, China

Yang-Sae Moon Kangwon National University, Korea

Yasuhiko Morimoto Hiroshima University, Japan

Yidong Li Beijing Jiaotong University, China

Yi-Dong Shen Chinese Academy of Sciences, China

Ying Zhang University of New South Wales, Australia

Yi-Ping Phoebe Chen La Trobe University, Australia

Yiu-ming Cheung Hong Kong Baptist University, Hong KongYong Guan Iowa State University, USA

Yonghong Peng University of Bradford, UK

Yue-Shi Lee Ming Chuan University, Taiwan

Zheng Chen Microsoft Research Asia, China

Zhenhui Li Pennsylvania State University, USA

Zhiyuan Chen University of Maryland, Baltimore County, USAZhongfei Zhang Binghamton University, USA

Zili Zhang Deakin University, Australia

External Reviewers

Ahsanul Haque University of Texas at Dallas, USA

Ameeta Agrawal York University, Canada

Anh Kim Nguyen Hanoi University of Science and Technology,

VietnamArnaud Soulet Université François Rabelais, Tours, FranceBhanukiran Vinzamuri Wayne State University, USA

Bin Fu University of Technology, Sydney, AustraliaBing Tian Dai Singapore Management University, SingaporeBudhaditya Saha Deakin University, Australia

Cam-Tu Nguyen Nanjing University, China

Cheng Long Hong Kong University of Science and Technology,

Hong KongChung-Hsien Yu University of Massachusetts Boston, USA

Chunming Liu University of Technology, Sydney, AustraliaDawei Wang University of Massachusetts Boston, USA

Dieu-Thu Le University of Trento, Italy

Dinusha Vatsalan Australian National University, Australia

Doan V Nguyen Japan Advanced Institute of Science and

Technology, JapanEmmanuel Coquery Université Lyon1, CNRS, France

Ettore Ritacco ICAR-CNR, Italy

Fan Jiang University of Manitoba, Canada

Fang Yuan Institute for Infocomm Research A*STAR,

SingaporeFangfang Li University of Technology, Sydney, AustraliaFernando Gutierrez University of Oregon, USA

Fuzheng Zhang University of Science and Technology of China,

ChinaGensheng Zhang University of Texas at Arlington, USA

Gianni Costa ICAR-CNR, Italy

Guan-Bin Chen National Cheng Kung University, Taiwan

Hao Wang University of Oregon, USA

Heidar Davoudi York University, Canada

Henry Lo University of Massachusetts Boston, USA

Ikumi Suzuki National Institute of Genetics, Japan

Jan Bazan University of Rzeszów, Poland

Jan Vosecky Hong Kong University of Science and Technology,

Hong KongJavid Ebrahimi University of Oregon, USA

Jianhua Yin Tsinghua University, China

Jianmin Li Tsinghua University, China

Jianpeng Xu Michigan State University, USA

Jing Ren Singapore Management University, SingaporeJinpeng Chen Beihang University, China

Jipeng Qiang University of Massachusetts Boston, USA

Joseph Paul Cohen University of Massachusetts Boston, USA

Junfu Yin University of Technology, Sydney, AustraliaJustin Sahs University of Texas at Dallas, USA

Kai-Ho Chan Hong Kong University of Science and Technology,

Hong KongKazuo Hara National Institute of Genetics, Japan

Ke Deng RMIT University, Australia

Kiki Maulana Adhinugraha Monash University, Australia

Kin-Long Ho Hong Kong University of Science and Technology,

Hong KongLan Thi Le Hanoi University of Science and Technology,

VietnamLei Zhu Huazhong University of Science and Technology,

ChinaLin Li Wuhan University of Technology, China

Linh Van Ngo Hanoi University of Science and Technology,

VietnamLoc Do Singapore Management University, SingaporeMaksim Tkachenko Singapore Management University, SingaporeMarc Plantevit Université de Lyon, France

Marian Scuturici INSA de Lyon, CNRS, France

Marthinus Christoffel du Plessis University of Tokyo, Japan

Md Anisuzzaman Siddique Hiroshima University, Japan

Min Xie Hong Kong University of Science and Technology,

Hong KongMing Yang Binghamton University, USA

Minh Nhut Nguyen Institute for Infocomm Research A*STAR,

SingaporeMohit Sharma University of Minnesota, USA

Morteza Zihayat York University, Canada

Mu Li University of Technology, Sydney, AustraliaNaeemul Hassan University of Texas at Arlington, USA

NhatHai Phan University of Oregon, USA

Nicola Barbieri Yahoo Labs, Spain

Nicolas Béchet Université de Bretagne Sud, France

Nima Shahbazi York University, Canada

Pakawadee Pengcharoen Hong Kong University of Science and Technology,

Hong KongPawel Gora University of Warsaw, Poland

Peiyuan Zhou Hong Kong Polytechnic University, Hong KongPeng Peng Hong Kong University of Science and Technology,

Hong KongPinghua Gong University of Michigan, USA

Qiong Fang Hong Kong University of Science and Technology,

Hong KongQuan Xiaojun Institute for Infocomm Research A*STAR,

SingaporeRiccardo Ortale ICAR-CNR, Italy

Sabin Kafle University of Oregon, USA

San Phyo Phyo Institute for Infocomm Research A*STAR,

SingaporeSang The Dinh Hanoi University of Science and Technology,

VietnamShangpu Jiang University of Oregon, USA

Shenlu Wang University of New South Wales, Australia

Shiyu Yang University of New South Wales, Australia

Show-Jane Yen Ming Chuan University, Taiwan

Shuangfei Zhai Binghamton University, USA

Simone Romano University of Melbourne, Australia

Sujatha Das Gollapalli Institute for Infocomm Research A*STAR,

SingaporeSwarup Chandra University of Texas at Dallas, USA

Syed K Tanbeer University of Manitoba, Canada

Tenindra Abeywickrama Monash University, Australia

Thanh-Son Nguyen Singapore Management University, SingaporeThin Nguyen Deakin University, Australia

Tiantian He Hong Kong Polytechnic University, Hong KongTianyu Kang University of Massachusetts Boston, USA

Trung Le Deakin University, Australia

Tuan M V Le Singapore Management University, SingaporeXiaochen Chen Google, USA

Xiaolin Hu Tsinghua University, China

Xin Li University of Science and Technology, ChinaXuhui Fan University of Technology, Sydney, AustraliaYahui Di University of Massachusetts Boston, USA

Yan Li Wayne State University, USA

Yang Jianbo Institute for Infocomm Research A*STAR,

SingaporeYang Mu University of Massachusetts Boston, USA

Yanhua Li University of Minnesota, USA

Yanhui Gu Nanjing Normal University, China

Yathindu Rangana Hettiarachchige Monash University, Australia

Yi-Yu Hsu National Cheng Kung University, Taiwan

Yingming Li Binghamton University, USA

Yu Zong West Anhui University, China

Zhiyong Chen Singapore Management University, SingaporeZhou Zhao Hong Kong University of Science and Technology,

Hong KongZongda Wu Wenzhou University, China

Social Networks and Social Media

Maximizing Friend-Making Likelihood for Social Activity

Organization 3Chih-Ya Shen, De-Nian Yang, Wang-Chien Lee, and Ming-Syan Chen

What Is New in Our City? A Framework for Event Extraction Using

Social Media Posts 16Chaolun Xia, Jun Hu, Yan Zhu, and Mor Naaman

Link Prediction in Aligned Heterogeneous Networks 33Fangbing Liu and Shu-Tao Xia

Scale-Adaptive Group Optimization for Social Activity Planning 45Hong-Han Shuai, De-Nian Yang, Philip S Yu, and Ming-Syan Chen

Influence Maximization Across Partially Aligned Heterogenous

Social Networks 58Qianyi Zhan, Jiawei Zhang, Senzhang Wang, Philip S Yu,

and Junyuan Xie

Multiple Factors-Aware Diffusion in Social Networks 70Chung-Kuang Chou and Ming-Syan Chen

Understanding Community Effects on Information Diffusion 82Shuyang Lin, Qingbo Hu, Guan Wang, and Philip S Yu

On Burst Detection and Prediction in Retweeting Sequence 96Zhilin Luo, Yue Wang, Xintao Wu, Wandong Cai, and Ting Chen

#FewThingsAboutIdioms: Understanding Idioms and Its Users

in the Twitter Online Social Network 108Koustav Rudra, Abhijnan Chakraborty, Manav Sethi, Shreyasi Das,

Niloy Ganguly, and Saptarshi Ghosh

Retweeting Activity on Twitter: Signs of Deception 122Maria Giatsoglou, Despoina Chatzakou, Neil Shah,

Christos Faloutsos, and Athena Vakali

Resampling-Based Gap Analysis for Detecting Nodes with High

Centrality on Large Social Network 135Kouzou Ohara, Kazumi Saito, Masahiro Kimura, and Hiroshi Motoda

Double Ramp Loss Based Reject Option Classifier 151Naresh Manwani, Kalpit Desai, Sanand Sasidharan,

and Ramasubramanian Sundararajan

Efficient Methods for Multi-label Classification 164Chonglin Sun, Chunting Zhou, Bo Jin, and Francis C.M Lau

A Coupled k-Nearest Neighbor Algorithm for Multi-label Classification 176Chunming Liu and Longbing Cao

Learning Topic-Oriented Word Embedding for Query Classification 188Hebin Yang, Qinmin Hu, and Liang He

Reliable Early Classification on Multivariate Time Series

with Numerical and Categorical Attributes 199Yu-Feng Lin, Hsuan-Hsu Chen, Vincent S Tseng, and Jian Pei

Distributed Document Representation for Document Classification 212Rumeng Li and Hiroyuki Shindo

Prediciton of Emergency Events: A Multi-Task Multi-Label Learning

Approach 226Budhaditya Saha, Sunil Kumar Gupta, and Svetha Venkatesh

Nearest Neighbor Method Based on Local Distribution for Classification 239Chengsheng Mao, Bin Hu, Philip Moore, Yun Su, and Manman Wang

Immune Centroids Over-Sampling Method for Multi-Class Classification 251Xusheng Ai, Jian Wu, Victor S Sheng, Pengpeng Zhao, Yufeng Yao,

and Zhiming Cui

Optimizing Classifiers for Hypothetical Scenarios 264Reid A Johnson, Troy Raeder, and Nitesh V Chawla

Repulsive-SVDD Classification 277Phuoc Nguyen and Dat Tran

Centroid-Means-Embedding: an Approach to Infusing Word Embeddings

into Features for Text Classification 289Mohammad Golam Sohrab, Makoto Miwa, and Yutaka Sasaki

Machine Learning

Collaborating Differently on Different Topics: A Multi-Relational Approach

to Multi-Task Learning 303Sunil Kumar Gupta, Santu Rana, Dinh Phung, and Svetha Venkatesh

Multi-Task Metric Learning on Network Data 317Chen Fang and Daniel N Rockmore

A Bayesian Nonparametric Approach to Multilevel Regression 330

Vu Nguyen, Dinh Phung, Svetha Venkatesh, and Hung H Bui

Learning Conditional Latent Structures from Multiple Data Sources 343Viet Huynh, Dinh Phung, Long Nguyen, Svetha Venkatesh,

and Hung H Bui

Collaborative Multi-view Learning with Active Discriminative Prior

for Recommendation 355Qing Zhang and Houfeng Wang

Online and Stochastic Universal Gradient Methods for Minimizing

Regularized Hölder Continuous Finite Sums in Machine Learning 369Ziqiang Shi and Rujie Liu

Context-Aware Detection of Sneaky Vandalism on Wikipedia Across

Multiple Languages 380Khoi-Nguyen Tran, Peter Christen, Scott Sanner, and Lexing Xie

Uncovering the Latent Structures of Crowd Labeling 392Tian Tian and Jun Zhu

Use Correlation Coefficients in Gaussian Process to Train Stable

ELM Models 405Yulin He, Joshua Zhexue Huang, Xizhao Wang, and Rana Aamir Raza

Local Adaptive and Incremental Gaussian Mixture for Online Density

Estimation 418Tianyu Qiu, Furao Shen, and Jinxi Zhao

Latent Space Tracking from Heterogeneous Data with an Application

for Anomaly Detection 429Jiaji Huang and Xia Ning

A Learning-Rate Schedule for Stochastic Gradient Methods to Matrix

Factorization 442Wei-Sheng Chin, Yong Zhuang, Yu-Chin Juan, and Chih-Jen Lin

Predicting Smartphone Adoption in Social Networks 472

Le Wu, Yin Zhu, Nicholas Jing Yuan, Enhong Chen, Xing Xie,

and Yong Rui

Discovering the Impact of Urban Traffic Interventions Using Contrast

Mining on Vehicle Trajectory Data 486Xiaoting Wang, Christopher Leckie, Hairuo Xie,

and Tharshan Vaithianathan

Locating Self-collection Points for Last-mile Logistics using Public

Transport Data 498Huayu Wu, Dongxu Shao, and Wee Siong Ng

A Stochastic Framework for Solar Irradiance Forecasting Using Condition

Random Field 511Jin Xu, Shinjae Yoo, Dantong Yu, Hao Huang, Dong Huang,

John Heiser, and Paul Kalb

Online Prediction of Chess Match Result 525Mohammad M Masud, Ameera Al-Shehhi, Eiman Al-Shamsi,

Shamma Al-Hassani, Asmaa Al-Hamoudi, and Latifur Khan

Learning of Performance Measures from Crowd-Sourced Data

with Application to Ranking of Investments 538Greg Harris, Anand Panangadan, and Viktor K Prasanna

Hierarchical Dirichlet Process for Tracking Complex Topical Structure

Evolution and its Application to Autism Research Literature 550Adham Beykikhoshk, Ognjen Arandjelovic´, Svetha Venkatesh,

and Dinh Phung

Automated Detection for Probable Homologous Foodborne Disease

Outbreaks 563Xiao Xiao, Yong Ge, Yunchang Guo, Danhuai Guo, Yi Shen,

Yuanchun Zhou, and Jianhui Li

Identifying Hesitant and Interested Customers for Targeted Social

Marketing 576Guowei Ma, Qi Liu, Le Wu, and Enhong Chen

Activity-Partner Recommendation 591Wenting Tu, David W Cheung, Nikos Mamoulis, Min Yang,

and Ziyu Lu

Iterative Use of Weighted Voronoi Diagrams to Improve Scalability

in Recommender Systems 605Joydeep Das, Subhashis Majumder, Debarshi Dutta,

and Prosenjit Gupta

Novel Methods and Algorithms

Principal Sensitivity Analysis 621Sotetsu Koyamada, Masanori Koyama, Ken Nakae, and Shin Ishii

SocNL: Bayesian Label Propagation with Confidence 633Yuto Yamaguchi, Christos Faloutsos, and Hiroyuki Kitagawa

An Incremental Local Distribution Network for Unsupervised Learning 646Youlu Xing, Tongyi Cao, Ke Zhou, Furao Shen, and Jinxi Zhao

Trend-Based Citation Count Prediction for Research Articles 659Cheng-Te Li, Yu-Jen Lin, Rui Yan, and Mi-Yen Yeh

Mining Text Enriched Heterogeneous Citation Networks 672Jan Kralj, Anita Valmarska, Marko Robnik-Šikonja, and Nada Lavracˇ

Boosting via Approaching Optimal Margin Distribution 684Chuan Liu and Shizhong Liao

o-HETM: An Online Hierarchical Entity Topic Model for News Streams 696Linmei Hu, Juanzi Li, Jing Zhang, and Chao Shao

Modeling User Interest and Community Interest in Microbloggings:

An Integrated Approach 708Tuan-Anh Hoang

Minimal Jumping Emerging Patterns: Computation and Practical

Assessment 722Bamba Kane, Bertrand Cuissart, and Bruno Crémilleux

Rank Matrix Factorisation 734Thanh Le Van, Matthijs van Leeuwen, Siegfried Nijssen,

and Luc De Raedt

An Empirical Study of Personal Factors and Social Effects on Rating

Prediction 747Zhijin Wang, Yan Yang, Qinmin Hu, and Liang He

Author Index 759

Opinion Mining and Sentiment Analysis

Emotion Cause Detection for Chinese Micro-Blogs Based on ECOCC

Model 3Kai Gao, Hua Xu, and Jiushuo Wang

Parallel Recursive Deep Model for Sentiment Analysis 15Changliang Li, Bo Xu, Gaowei Wu, Saike He, Guanhua Tian,

and Yujun Zhou

Sentiment Analysis in Transcribed Utterances 27Nir Ofek, Gilad Katz, Bracha Shapira, and Yedidya Bar-Zev

Rating Entities and Aspects Using a Hierarchical Model 39Xun Wang, Katsuhito Sudoh, and Masaaki Nagata

Sentiment Analysis on Microblogging by Integrating Text and Image

Features 52Yaowen Zhang, Lin Shang, and Xiuyi Jia

TSum4act: A Framework for Retrieving and Summarizing Actionable

Tweets during a Disaster for Reaction 64Minh-Tien Nguyen, Asanobu Kitamoto, and Tri-Thanh Nguyen

Clustering

Evolving Chinese Restaurant Processes for Modeling Evolutionary

Traces in Temporal Data 79Peng Wang, Chuan Zhou, Peng Zhang, Weiwei Feng, Li Guo,

and Binxing Fang

Small-Variance Asymptotics for Bayesian Nonparametric Models

with Constraints 92Cheng Li, Santu Rana, Dinh Phung, and Svetha Venkatesh

Spectral Clustering for Large-Scale Social Networks via a Pre-Coarsening

Sampling Based NystrÖm Method 106Ying Kang, Bo Yu, Weiping Wang, and Dan Meng

pcStream: A Stream Clustering Algorithm for Dynamically Detecting

and Managing Temporal Contexts 119Yisroel Mirsky, Bracha Shapira, Lior Rokach, and Yuval Elovici

Clustering Over Data Streams Based on Growing Neural Gas 134Mohammed Ghesmoune, Mustapha Lebbah, and Hanene Azzag

Computing and Mining ClustCube Cubes Efficiently 146Alfredo Cuzzocrea

Outlier and Anomaly Detection

Contextual Anomaly Detection Using Log-Linear Tensor Factorization 165Alpa Jayesh Shah, Christian Desrosiers, and Robert Sabourin

A Semi-Supervised Framework for Social Spammer Detection 177Zhaoxing Li, Xianchao Zhang, Hua Shen, Wenxin Liang,

Christos Faloutsos, and Athena Vakali

An Embedding Scheme for Detecting Anomalous Block Structured

Graphs 215Lida Rashidi, Sutharshan Rajasegarar, and Christopher Leckie

A Core-Attach Based Method for Identifying Protein Complexes

in Dynamic PPI Networks 228Jiawei Luo, Chengchen Liu, and Hoang Tu Nguyen

Mining Uncertain and Imprecise Data

Mining Uncertain Sequential Patterns in Iterative MapReduce 243Jiaqi Ge, Yuni Xia, and Jian Wang

Quality Control for Crowdsourced POI Collection 255Shunsuke Kajimura, Yukino Baba, Hiroshi Kajino,

and Hisashi Kashima

Towards Efficient Sequential Pattern Mining in Temporal Uncertain

Databases 268Jiaqi Ge, Yuni Xia, and Jian Wang

Preference-Based Top-k Representative Skyline Queries on Uncertain

Databases 280

Ha Thanh Huynh Nguyen and Jinli Cao

Cluster Sequence Mining: Causal Inference with Time and Space

Proximity under Uncertainty 293Yoshiyuki Okada, Ken-ichi Fukui, Koichi Moriyama,

and Masayuki Numao

Achieving Accuracy Guarantee for Answering Batch Queries

with Differential Privacy 305Dong Huang, Shuguo Han, and Xiaoli Li

Mining Temporal and Spatial Data

Automated Classification of Passing in Football 319Michael Horton, Joachim Gudmundsson, Sanjay Chawla,

and Joël Estephan

Stabilizing Sparse Cox Model Using Statistic and Semantic Structures

in Electronic Medical Records 331Shivapratap Gopakumar, Tu Dinh Nguyen, Truyen Tran,

Dinh Phung, and Svetha Venkatesh

Predicting Next Locations with Object Clustering and Trajectory

Clustering 344Meng Chen, Yang Liu, and Xiaohui Yu

A Plane Moving Average Algorithm for Short-Term Traffic Flow

Prediction 357Lei Lv, Meng Chen, Yang Liu, and Xiaohui Yu

Recommending Profitable Taxi Travel Routes Based on Big Taxi

Trajectories Data 370Wenxin Yang, Xin Wang, Seyyed Mohammadreza Rahimi,

and Jun Luo

Semi Supervised Adaptive Framework for Classifying Evolving Data

Stream 383Ahsanul Haque, Latifur Khan, and Michael Baron

Feature Extraction and Selection

Cost-Sensitive Feature Selection on Heterogeneous Data 397Wenbin Qian, Wenhao Shu, Jun Yang, and Yinglong Wang

A Feature Extraction Method for Multivariate Time Series Classification

Using Temporal Patterns 409Pei-Yuan Zhou and Keith C.C Chan

Scalable Outlying-Inlying Aspects Discovery via Feature Ranking 422Nguyen Xuan Vinh, Jeffrey Chan, James Bailey, Christopher Leckie,

Kotagiri Ramamohanarao, and Jian Pei

A DC Programming Approach for Sparse Optimal Scoring 435Hoai An Le Thi and Duy Nhat Phan

Graph Based Relational Features for Collective Classification 447Immanuel Bayer, Uwe Nagel, and Steffen Rendle

A New Feature Sampling Method in Random Forests for Predicting

High-Dimensional Data 459Thanh-Tung Nguyen, He Zhao, Joshua Zhexue Huang, Thuy Thi Nguyen,

and Mark Junjie Li

Mining Heterogeneous, High Dimensional, and Sequential Data

Seamlessly Integrating Effective Links with Attributes for Networked

Data Classification 473Yangyang Zhao, Zhengya Sun, Changsheng Xu, and Hongwei Hao

Clustering on Multi-source Incomplete Data via Tensor Modeling

and Factorization 485Weixiang Shao, Lifang He, and Philip S Yu

Locally Optimized Hashing for Nearest Neighbor Search 498Seiya Tokui, Issei Sato, and Hiroshi Nakagawa

Do-Rank: DCG Optimization for Learning-to-Rank in Tag-Based Item

Recommendation Systems 510Noor Ifada and Richi Nayak

Efficient Discovery of Recurrent Routine Behaviours in Smart Meter Time

Series by Growing Subsequences 522Jin Wang, Rachel Cardell-Oliver, and Wei Liu

Convolutional Nonlinear Neighbourhood Components Analysis for Time

Series Classification 534

Yi Zheng, Qi Liu, Enhong Chen, J Leon Zhao, Liang He,

and Guangyi Lv

Entity Resolution and Topic Modelling

Clustering-Based Scalable Indexing for Multi-party Privacy-Preserving

Record Linkage 549Thilina Ranbaduge, Dinusha Vatsalan, and Peter Christen

Efficient Interactive Training Selection for Large-Scale Entity Resolution 562Qing Wang, Dinusha Vatsalan, and Peter Christen

Unsupervised Blocking Key Selection for Real-Time Entity Resolution 574Banda Ramadan and Peter Christen

Incorporating Probabilistic Knowledge into Topic Models 586Liang Yao, Yin Zhang, Baogang Wei, Hongze Qian, and Yibing Wang

Learning Focused Hierarchical Topic Models with Semi-Supervision

in Microblogs 598Anton Slutsky, Xiaohua Hu, and Yuan An

Predicting Future Links Between Disjoint Research Areas Using

Heterogeneous Bibliographic Information Network 610Yakub Sebastian, Eu-Gene Siew, and Sylvester Olubolu Orimaye

Itemset and High Performance Data Mining

CPT+: Decreasing the Time/Space Complexity of the Compact Prediction

Tree 625Ted Gueniche, Philippe Fournier-Viger, Rajeev Raman,

and Vincent S Tseng

Mining Association Rules in Graphs Based on Frequent Cohesive

Itemsets 637Tayena Hendrickx, Boris Cule, Pieter Meysman, Stefan Naulaerts,

Kris Laukens, and Bart Goethals

Mining High Utility Itemsets in Big Data 649Ying Chun Lin, Cheng-Wei Wu, and Vincent S Tseng

Decomposition Based SAT Encodings for Itemset Mining Problems 662Said Jabbour, Lakhdar Sais, and Yakoub Salhi

A Comparative Study on Parallel LDA Algorithms in MapReduce

Framework 675Yang Gao, Zhenlong Sun, Yi Wang, Xiaosheng Liu, Jianfeng Yan,

and Jia Zeng

Distributed Newton Methods for Regularized Logistic Regression 690Yong Zhuang, Wei-Sheng Chin, Yu-Chin Juan, and Chih-Jen Lin

Recommendation

Coupled Matrix Factorization Within Non-IID Context 707Fangfang Li, Guandong Xu, and Longbing Cao

Complementary Usage of Tips and Reviews for Location

Recommendation in Yelp 720Saurabh Gupta, Sayan Pathak, and Bivas Mitra

Coupling Multiple Views of Relations for Recommendation 732Bin Fu, Guandong Xu, Longbing Cao, Zhihai Wang, and Zhiang Wu

Pairwise One Class Recommendation Algorithm 744Huimin Qiu, Chunhong Zhang, and Jiansong Miao

RIT: Enhancing Recommendation with Inferred Trust 756Guo Yan, Yuan Yao, Feng Xu, and Jian Lu

Author Index 769

Social Networks and Social Media

for Social Activity Organization

Chih-Ya Shen1(B), De-Nian Yang2, Wang-Chien Lee3, and Ming-Syan Chen1,2

1 Research Center for Information Technology Innovation,

Academia Sinica, Taipei, Taiwan

{chihya,mschen}@citi.sinica.edu.tw

2 Institute of Information Science, Academia Sinica, Taipei, Taiwan

dnyang@iis.sinica.edu.tw

3 Department of Computer Science and Engineering,

The Pennsylvania State University, University Park, USA

wlee@cse.psu.edu

Abstract The social presence theory in social psychology suggests that

computer-mediated online interactions are inferior to face-to-face, person interactions In this paper, we consider the scenarios of organiz-ing in person friend-making social activities via online social networks(OSNs) and formulate a new research problem, namely, Hop-boundedMaximum Group Friending (HMGF), by modeling both existing friend-ships and the likelihood of new friend making To find a set of atten-dees for socialization activities, HMGF is unique and challenging due

in-to the interplay of the group size, the constraint on existing friendshipsand the objective function on the likelihood of friend making We provethat HMGF is NP-Hard, and no approximation algorithm exists unless

P = N P We then propose an error-bounded approximation algorithm

to efficiently obtain the solutions very close to the optimal solutions Weconduct a user study to validate our problem formulation and performextensive experiments on real datasets to demonstrate the efficiency andeffectiveness of our proposed algorithm

of these activities shows that OSNs have been widely used as a convenient meansfor initiating real-life activities among friends

1 http://www.skout.com/

2 http://newsroom.fb.com/products/

3 http://www.meetup.com/about/

c

 Springer International Publishing Switzerland 2015

T Cao et al (Eds.): PAKDD 2015, Part I, LNAI 9077, pp 3–15, 2015.

On the other hand, to help users expand their circles of friends in thecyberspace, friend recommendation services have been provided in OSNs to sug-gest candidates to users who may likely become mutual friends in the future.Many friend recommendation services employ link prediction algorithms, e.g.,[10,11], to analyze the features, similarity or interaction patterns of users inorder to derive potential future friendship between some users By leveragingthe abundant information in OSNs, link prediction algorithms show high accu-racy for recommending online friends in OSNs.

As social presence theory [16] in social psychology suggests, computer-mediatedonline interactions are inferior to face-to-face, in-person interactions, off-line friend-making activities may be favorable to their on-line counterparts in cyberspace.Therefore, in this paper, we consider the scenarios of organizing face-to-face friend-making activities via OSN services Notice that finding socially cohesive groups ofparticipants is essential for maintaining good atmosphere for the activity More-over, the function of making new friends is also an important factor for the success ofsocial activities, e.g., assigning excursion groups in conferences, inviting attendees

to housewarming parties, etc Thus, for organizing friend-making social activities,both activity organization and friend recommendation services are fundamental.However, there is a gap between existing activity organization and friend recom-mendation services in OSNs for the scenarios under consideration Existing activityorganization approaches focus on extracting socially cohesive groups from OSNsbased on certain cohesive measures, density, diameter, of social networks or otherconstraints, e.g., time, spatial distance, and interests, of participants [5 8] On the

other hand, friend recommendation services consider only the existing friendships

to recommend potential new friends for an individual (rather than finding a group

of people for engaging friend-making) We argue that in addition to themes of mon interests, it is desirable to organize friend-making activities by mixing the

com-”potential friends”, who may be interested in knowing each other (as indicated by alink prediction algorithm), with existing friends (as lubricators) To the best knowl-edge of the authors, the following two important factors, 1) the existing friendshipamong attendees, and 2) the potential friendship among attendees, have not beenconsidered simultaneously in existing activity organization services To bridge thegap, it is desirable to propose a new activity organization service that carefullyaddresses these two factors at the same time

In this paper, we aim to investigate the problem of selecting a set of didate attendees from the OSN by considering both the existing and potentialfriendships among the attendees To capture the two factors for activity organi-zation, we propose to include the likelihood of making new friends in the socialnetwork As such, we formulate a new research problem to find groups withtight social relationships among existing friends and potential friends (i.e., whoare not friends yet) Specifically, we model the social network in the OSN as

can-a heterogeneous socican-al grcan-aph G = (V, E, R) with edge weight w : R → (0, 1],

where V is the set of individuals, E is the set of friend edges, and R is the set

of potential friend edges (or potential edges for short) Here a friend edge (u, v) denotes that individuals u and v are mutual friends, while a potential edge [u  , v ]

f

b c a

(a) Input Graph G.

g f a

e

0.7

0.9 0.8

0.4 0.5

0.7

0.9 0.8

(d) H3

Fig 1 Illustrative Example

indicates that individuals u  and v  are likely to become friends (the edge weight

w[u  , v ] quantifies the likelihood) The potential edges and the correspondingedge weights can be obtained by employing a link prediction algorithm in friendrecommendation

Given a heterogeneous social graph G = (V, E, R) as described above, we formulate a new problem, namely, Hop-bounded Maximum Group Friending

(HMGF), to find a group that 1) maximizes the likelihood of making new friends

among the group, i.e., the group has the highest ratio of total potential edgeweight to group size, 2) ensures that the social tightness, i.e., hop count on

friend edges in G between each pair of individuals is small, and 3) is a

suffi-ciently large group, i.e., too small a group may not work well for socializationactivities

Figure 1 illustrates the social graph and the interplay of the above factors.Figure1(a)shows a social graph, where a dash line, e.g., [a, b] with weight 0.6, is

a potential edge and a solid line, e.g., (c, d), is a friend edge Figure 1(b)shows

a group H1:{a, e, f, g} which has many potential edges and thus a high total

weight However, not all the members of this group have common friends associal lubricators Figure1(c)shows a group H2:{c, d, f, g} tightly connected by

friend edges While H2 may be a good choice for gathering of close friends, thegoal of friend-making in socialization activities is missed Finally, Figure 1(d)

shows H3:{d, e, f, g} which is a better choice than H1 and H2 for socialization

activities because each member of H3 is within 2 hops of another member via

friend edges in G Moreover, the average potential edge weight among them is

high, indicating members are likely to make some new friends

Processing HMGF to find the best solution is very challenging because thereare many important factors to consider, including hop constraint, group size andthe total weight of potential edges in a group Indeed, we prove that HMGF is anNP-Hard problem with no approximation algorithm Nevertheless, we prove that

if the hop constraint can be slightly relaxed to allow a small error, there exists a3-approximation algorithm for HMGF Theoretical analysis and empirical resultsshow that our algorithm can obtain good solutions efficiently

The contributions made in this study are summarized as follows

– For socialization activity organization, we propose to model the existingfriendship and the potential friendship in a heterogeneous social graph andformulate a new problem, namely, Hop-bounded Maximum Group Friending(HMGF), for finding suitable attendees To our best knowledge, HMGF is

the first problem that considers these two important relationships betweenattendees for activity organization.

– We prove that HMGF is NP-Hard and there exists no approximation

rithm for HMGF unless P = N P We then propose an approximation

algo-rithm, called MaxGF, with a guaranteed error bound for solving HMGFefficiently

– We conduct a user study on 50 users to validate our argument for ing both existing and potential friendships in activity organization We alsoperform extensive experiments on real datasets to evaluate the proposedalgorithm Experimental results manifest that HMGF can obtain solutionsvery close to the optimal ones, very efficiently

consider-The rest of this paper is organized as follows Section2 formulates HMGFand proves it NP-Hard with no approximation algorithm Section3reviews therelated works, and Section 4 details the algorithm design Section 5 reports auser study and experimental results Section6 concludes this paper

Based on the description of heterogeneous social graph described earlier, here we

formulate the Hop-bounded Maximum Group Friending (HMGF) tackled in this paper Given two individuals u and v, let d E (u, v) be the shortest path between

u and v via friend edges in G Moreover, given H ⊆ G, let w(H) denote the total

weight of potential edges in H and let average weight, σ(H) = w(H) |H| denote the

average weight of potential edges connected to each individual in H4 HMGF isformulated as follows

Problem: Hop-bounded Maximum Group Friending (HMGF).

Given: Social network G = (V, E, R), hop constraint h, and size constraint p.

|H| ≥ p and d E (u, v) ≤ h, ∀u, v ∈ H.

Efficient processing of HMGF is very challenging due to the following reasons:

1) The interplay of the total weight w(H) and the size of H To maximize σ(H), finding a small H may not be a good choice because the number of edges in

a small graph tends to be small as well On the other hand, finding a large H (which usually has a high w(H)) may not lead to an acceptable σ(H), either.

Therefore, the key is to strike a good balance between the graph size|H| and the

total weight w(H) 2) HMGF includes a hop constraint (say h = 2) on friend

edges to ensure that every pair of individuals is not too distant socially from

each other However, selecting a potential edge [u, v] with a large weight w[u, v] may not necessarily satisfy the hop constraint, i.e., d E (u, v) > h which is defined

based on existing friend edges In this case, it may not always be a good strategy

to prioritize on large-weight edges in order to maximize σ(H), especially when

u and v do not share a common friend nearby via the friend edges.

4 Note that σ(H) = 0 if H = ∅.

In the following, we prove that HMGF is NP-Hard and not approximable

within any factor In other words, there exists no approximation algorithm forHMGF

Theorem 1 HMGF is NP-Hard and there is no approximation algorithm for

HMGF unless P = N P

Proof Due to the space constraints, we prove this theorem in the full version of

this paper (available online [1])

The above theorem manifests that HMGF has no approximation algorithm

Nevertheless, we show that HMGF becomes approximable if a small error h is

allowed in the hop constraint More specifically, in Section 4, we first propose

an error-bounded approximation algorithm for HMGF, which returns a solution

with guaranteed σ(H), while d E (u, v) for any two vertices u and v in H may exceed h but is always bounded by 2h Afterward, we present a post-processing

procedure to tailor the solution for satisfying the hop constraint

Extracting dense subgraphs or social cohesive groups among social networks is

a natural way for selecting a set of close friends for a gathering Various socialcohesive measures have been proposed for finding dense social subgraphs, e.g.,diameter [2], density [3], clique and its variations [4] Although these social cohe-sive measures cover a wide range of application scenarios, they focus on deriv-ing groups based only on existing friendship in the social network In contrast,the HMGF studied in this paper aims to extract groups by considering boththe existing and potential friendships for socialization activities Therefore, theexisting works mentioned above cannot be directly applied to HMGF tackled inthis paper

Research on finding a set of attendees for activities based on the social ness among existing friends [5 9] have been reported in the literature Social-Temporal Group Query [5] checks the available times of attendees to find thesocial cohesive group with the most suitable activity time Geo-Social GroupQuery [6,7] extracts socially tight groups while considering certain spatial prop-erties The willingness optimization for social group problem in [8] selects a set

tight-of attendees for an activity while maximizing their willingness to participate.Finally, [9] finds a set of compatible members with tight social relationships

in the collaboration network Although these works find suitable attendees foractivities based on existing friendship among the attendees, they ignore the like-lihood of making new friends among the attendees Therefore, these works maynot be suitable for socialization activities discussed in this paper

Link prediction analyzes the features, similarity or interaction patterns amongindividuals in order to recommend possible friends to the users [10–14] Link pre-diction algorithms employ different approaches including graph-topological fea-tures, classification models, hierarchical probabilistic model, and linear algebraicmethods These works show good prediction accuracy for friend recommendation

in social networks In this paper, to estimate the likelihood of how individuals maypotentially become friends in the future, we employ link prediction algorithms forderiving the potential edges among the individuals.

To the best knowledge of the authors, there exists no algorithm for activityorganization that considers both the existing friendship and the likelihood ofmaking new friends when selecting activity attendees The HMGF studied in thispaper examines the social tightness among existing friends and the likelihood ofbecoming friends for non-friend attendees We envisage that our research resultcan be employed in various social network applications for activity organization

To tackle HMGF, a naive approach is to enumerate all possible combinations of

vertices, and extracts the subgraph H with the maximum σ(H) following the hop

and group size constraints However, this approach is computationally expensiveand thus not applicable for a large-scale social network To efficiently answer

HMGF, we propose an algorithm, called MaxGF, which is a 3-approximation algorithm with a guaranteed error bound h MaxGF limits the search space of

candidate solutions by dividing the graph into different hop-bounded subgraphssuch that their sizes are much smaller than|V | Then, it employs a greedy app-

roach on the hop-bounded subgraphs to iteratively remove the vertices that are

inclined to generate a small σ(H) Specifically, we define the incident weight of a vertex v in an induced subgraph H ⊆ G as τ H (v), where τ H (v) =

u∈H w[v, u],

i.e., the incident weight of v is the total weight of the potential edges incident

to v in H By carefully examining the incident weights of the vertices, we can

remove from the hop-bounded subgraph those vertices that contribute no gain

in the objective function Moreover, we propose an effective pruning strategy fortrimming redundant search Finally, a post-processing procedure is proposed toensure that the returned solution follows the hop constraint

The pseudo code of MaxGF is presented in Algorithm 1 Basically, to obtainthe hop-bounded subgraphs, MaxGF sorts the vertices in terms of their incident

weights and iteratively selects a vertex v with the maximum incident weight from G as a reference vertex A hop-bounded subgraph H v is constructed from

v by including every vertex u with at most h hops from v on the friend edges,

i.e., H v ={u|d E (u, v) ≤ h} Moreover, if |H v | < p, it is no longer necessary to

examine H v because any subgraph in H vwill never be a feasible solution due tothe size constraint Therefore, redundant search space is effectively pruned

In addition, another pruning condition is also proposed to further prune the

resulted subgraph H v Let S AP X denote the best solution obtained so far If

half of the maximum incident weight among the vertices u in H v , i.e., (1/2) ·

maxu∈H v τ H v (u), does not exceed σ(S AP X), there will never be any solution

Algorithm 1 MaxGF

Input: Social graph G = (V, E, R), hop constraint h, and size constraint p

2: while U = ∅ do

3: v ← arg max u∈U τ G (u), U ← U − {v}

4: let H v be the induced subgraph of G with vertices as {u|d E G (u, v) ≤ h}

2· max u∈H v τ H v (u) ≤ σ(S AP X) holds, there exists no subgraph in

H v with the average weight larger than σ(S AP X ), and H v can be pruned

Next, MaxGF starts to find the solution in H v with the maximized averageweight, which includes|H v | steps Let S i+1 denote the subgraph after removing

a vertex ˆv i from S i in step i That is, we set S1= H v initially, and at each step i afterwards, S i+1 is the subgraph S i −{ˆv i } During each step i, ˆv iis selected as the

vertex which has the lowest incident weight in S i, i.e., ˆv i = arg minu∈S i τ i (u).

This is based on the intuition that excluding vertices with low incident weights

is more inclined to increase the average weight of the the remaining subgraph.Then, ˆv i and its incident potential edges are removed from S iand the remaining

graph is S i+1 Then, S i+1 is processed in the next step i+1 The above procedure ends until S i is empty

To maximize the objective function σ(H) = w(H) |H| , after a hop-bounded

sub-graph H v is processed, S ∗ is extracted as the subgraph S i with the maximum

σ(S i ) in H v where |S i | ≥ p If σ(S ∗ ) > σ(S AP X ), we replace S AP X with S ∗.

Then, we continue to extract the next vertex v  for examining the corresponding

hop-bounded subgraph H v  until all vertices have been examined Afterward,

a post-processing procedure (detailed in Section 4.3) is employed on the best

solution obtained in the algorithm, i.e., S AP X, to ensure that the hop constraint

is satisfied and to further maximize σ(S AP X ) Finally, S AP X is output as thesolution

4.2 Theoretical Bound

In the following, given the hop-bounded subgraph H v, we first prove that there

exists a subgraph F ⊆ H v such that 3· w(F ) is an upper bound of the total

potential edge weight of the optimal solution to the HMGF instance on H v

Then, we prove that for each H v , the average weight of S ∗ obtained in the

algorithm, i.e., σ(S ∗), is at least 13 the average weight of the optimal solution

of HMGF on H v Finally, based on the properties of the hop-bounded subgraph

and S AP X, we prove that the proposed algorithm is a 3-approximation algorithmwith guaranteed error bound to HMGF

Let S v OP T denote the optimal solution of the HMGF instance on H v with

σ(S v OP T ) > 0, we first prove that the largest subgraph F in H v , where τ F (u) ≥

2

3σ(S v OP T),∀u ∈ F , is not an empty graph.

3σ(S v OP T ), ∀u ∈ F ,

is not an empty graph.

Proof The proof is presented in the online version [1]

With the existence of F proven above, we now derive an upper bound of the total potential edge weight of S v OP T , i.e., w(S v OP T ), according to w(F ).

Proof The proof is presented in the online version [1]

Then, with the properties derived above, we turn our attention to analyzing

MaxGF proposed in this section In MaxGF, given H v and when we are tively extracting ˆv i which has the minimum incident weight in S i, if ˆv i is thefirst extracted vertex such that ˆv i ∈ F (i.e., step i is the earliest step such that

itera-ˆi ∈ F ), then we have the following lemma.

ˆi from S i is in F , then τ S i (u) ≥ 2

3σ(S v OP T ), ∀u ∈ S i Moreover, F = S i Proof The proof is presented in the online version [1]

We combine the results obtained above, and derive the bound on σ(S ∗), where

S ∗ is the group S i which has the maximum σ(S i ) among all S i with |S i | ≥ p

obtained by MaxGF in H v Please note that Lemma 3 proves that during thesteps of extracting ˆv i from S i, there exists ˆv i with τ S i(ˆv i)≥2

Proof The proof is presented in the online version [1]

Finally, let S OP T denote the optimal solution of HMGF on G, the following theorem proves that the solution obtained by MaxGF, i.e., S AP X , has σ(S AP X)

at least 1· σ(S OP T ), and the error is bounded by h.

Theorem 3 MaxGF returns the solution S AP X with σ(S AP X)≥ σ(S OP T)

d E (u, v) ≤ 2 · h, ∀u, v ∈ S AP X .

Proof The proof is presented in the online version [1]

A post-processing procedure is designed to tailor S AP X for meeting the hopconstraint and further maximizing the average weight More specifically, given

S AP X obtained in the algorithm, we first define the notion of boundary vertices.

A vertex u in S AP X is a boundary vertex if there exists at least one other vertex

v in S AP X such that the shortest path from u to v via friend edges contains more than h edges Let B denote the set of boundary vertices MaxGF includes the

following adjustment steps in the post-processing procedure 1) Expand: a vertex

v ∈ (V \S AP X ) can be added into S AP X if adding v does not increase |B| and

increases σ(S AP X ) We give priority to the v which maximizes σ(S AP X ∪{v}) 2)

Shrink: given a boundary vertex u ∈ B, u can be safely removed if after removing

u from S AP X,|B| decreases but σ(S AP X ) does not We give priority to the u that

maximizes σ(S AP X −{u}) Please note that the above post-processing procedure

minimizes maxu,v∈S AP X d E (u, v) while increasing σ(S AP X) Therefore, after postprocessing, the performance and error bounds in Theorem3 still hold

The detailed analysis is presented in the online version of this paper [1]

We implement HMGF in Facebook and invite 50 users to participate in our userstudy Each user, given 12 test cases of HMGF using her friends in Facebook asthe input graph, is asked to solve the HMGF cases, and compare her results withthe solutions obtained by MaxGF In addition to the user study, we evaluate theperformance of MaxGF on two real social network datasets, i.e., FB [15] andthe MS dataset from KDD Cup 20135 The FB dataset is extracted from Face-book with 90K vertices, and MS is a co-author network with 1.7M vertices Weextract the friend edges from these datasets and identify the potential edges with

a link prediction algorithm [11] The weight of a potential edge is ranged within(0,1] Moreover, we compare MaxGF with two algorithms, namely, Baseline andDkS [3] Baseline finds the optimal solution of HMGF by enumerating all the

subgraphs satisfying the constraints, while DkS is an O( |V |1/3)-approximation

algorithm for finding a p-vertex subgraph H ⊆ G with the maximum density

on E ∪ R without considering the potential edges and the hop constraint The

algorithms are implemented in an IBM 3650 server with Quadcore Intel X54503.0 GHz CPUs We measure 30 samples in each scenario In the following, Fea-Ratio and ObjRatio respectively denote the ratio of feasibility (i.e., the portion

5 https://www.kaggle.com/c/kdd-cup-2013-author-paper-identification-challenge/

data