Broad Learning Through Fusions: An Application on Social Networks

424 Pages·2019·10.09 MB·English

by  Jiawei Zhang, Philip S. Yu

Jiawei Zhang · Philip S. Yu
Broad Learning
Through
Fusions
An Application on Social Networks
Broad Learning Through Fusions
Jiawei Zhang • Philip S. Yu
Broad Learning Through Fusions
An Application on Social Networks
123
JiaweiZhang PhilipS.Yu
DepartmentofComputerScience DepartmentofComputerScience
FloridaStateUniversity UniversityofIllinois
Tallahassee,FL,USA Chicago,IL,USA
ISBN978-3-030-12527-1 ISBN978-3-030-12528-8 (eBook)
https://doi.org/10.1007/978-3-030-12528-8
©SpringerNatureSwitzerlandAG2019
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartofthematerialisconcerned,
specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,broadcasting,reproductiononmicrofilmsorin
anyotherphysicalway,andtransmissionorinformationstorageandretrieval,electronicadaptation,computersoftware,orby
similarordissimilarmethodologynowknownorhereafterdeveloped.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublicationdoesnotimply,
evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevantprotectivelawsandregulationsand
thereforefreeforgeneraluse.
Thepublisher,theauthors,andtheeditorsaresafetoassumethattheadviceandinformationinthisbookarebelievedtobetrue
andaccurateatthedateofpublication.Neitherthepublishernortheauthorsortheeditorsgiveawarranty,expressorimplied,
withrespecttothematerialcontainedhereinorforanyerrorsoromissionsthatmayhavebeenmade.Thepublisherremains
neutralwithregardtojurisdictionalclaimsinpublishedmapsandinstitutionalaffiliations.
ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG.
Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland
Tomydearparents
YinheZhangandZhulanLiu.
JiaweiZhang
Tomyfamily.
PhilipS.Yu
Preface
Thistextbookiswrittenforthereaderswhoareinterestedinbroadlearning,especiallyininformation
fusionandknowledgediscoveryacrossmultiplefusedinformationsources.Broadlearningisageneral
learning problem, which can be studied in various disciplines. Meanwhile, to illustrate the problem
settings and the learning algorithms more clearly, this book uses the online social network as an
example. To make the textbook self-contained, the book also provides an overview of necessary
backgroundknowledgeforthereaders.Ifitisthefirsttimeforthereaderstoreadatextbookrelated
tobroadlearning,machinelearning,datamining,andsocialnetworkmining,thereaderswillfindthis
booktobeveryeasytofollow.
OverviewoftheBook
There are 12 chapters in this textbook, which are divided into four main parts: Part I covers
Chaps.1–3,whichintroducetheoverviewofbroadlearning,machinelearning,andsocialnetworks
for the readers; Part II covers Chaps. 4–6, which include the existing social network alignment
problems and algorithms; Part III covers Chaps. 7–11, which provide a comprehensive description
about the recent social media mining problems across multiple information sources; Part IV covers
Chap. 12, which indicates some potential future development of broad learning for the readers.
ReadersandinstructorscanusethistextbookaccordingtotheguidanceprovidedinChap.1.
Except Chap. 12, each chapter also has ten exercises for the readers. The exercise questions are
divided into three levels: easy, medium, and hard, which are on the basic concepts, theorem proofs,
algorithm details, as well as exercises on algorithm implementations. Some of the exercises can be
usedintheafter-classhomework,andsomecanbeusedasthecourseprojectsinstead.Theinstructors
candeterminehowtousetheexercisesaccordingtotheirdifficultylevels,aswellastheneedsofthe
courses.
Broad learning is a very large topic, and we cannot cover all the materials in this textbook. For
the readers who want to further explore other related areas, at the end of each chapter, we provide
a section about the bibliography notes. The readers can refer to the cited articles for more detailed
informationaboutthematerialstoyourinterests.
Acknowledgments
Thisbookwouldnothavebeenpossiblewithoutmanycontributorswhosenamesdidnotmakeittothe
cover.Wewillmentionthemhereaccordingtotheappearanceorderoftheircontributedworksinthis
book. The active network alignment algorithm introduced in Sect. 6.4 is based on the collaborative
work with Junxing Zhu (National University of Defense Technology) and the work with Yuxiang
vii
viii Preface
Ren (Florida State University). The large-scale network synergistic community detection algorithm
introducedinSect.8.5isbasedonthecollaborativeworkwithSongchangJin(NationalUniversityof
DefenseTechnology).TheinformationdiffusionalgorithmintroducedinSect.9.4andviralmarketing
algorithmintroducedinSect.10.4arebasedonthecollaborativeworkswithQianyiZhan(Jiangnan
University).
Thisbookhasbenefittedfromthesignificantfeedbacksfromourstudents,friends,andcolleagues.
We would like to thank many people who help to read and review the book which greatly improve
both the organization of the book and the detailed contents covered in the book. We would like to
thank Lin Meng (Florida State University) for helping review Chaps. 3, 6, 7, and 11; Yuxiang Ren
(FloridaStateUniversity)forreviewingChaps.2,4,5,and6;andQianyiZhan(JiangnanUniversity)
forreviewingChaps.9and10.
Jiawei would like to thank his long-term collaborators, including (sorted according to their last
names)CharuC.Aggarwal,YiChang,JianhuiChen,BowenDong,YanjieFu,LifangHe,QingboHu,
SongchangJin,XiangnanKong,MoyinLi,TaisongLi,KunpengLiu,YeLiu,YuanhuaLv,Guixiang
Ma, Xiao Pan, Weixiang Shao, Chuan Shi, Weiwei Shi, Lichao Sun, Pengyang Wang, Sen-Zhang
Wang, Yang Yang, Chenwei Zhang, Qianyi Zhan, Lei Zheng, Shi Zhi, Junxing Zhu, and Zhu-Hua
Zhou.JiaweialsowantstothankhisPhDadvisorPhilipS.Yuforhisguidanceduringtheearlyyears
asaresearcherandthemembersofhisIFMLab(InformationFusionandMiningLaboratory).Finally,
Jiaweiwouldliketothankhisrespectedparents,YinheZhangandZhulanLiu,fortheirselflesslove
andsupport.Thebookgrabssomuchtimethatshouldbespentinaccompanyingthem.
Philip would like to thank his collaborators, including past and current members of his BDSC
(BigDataandSocialComputing)LabatUIC.
SpecialthanksareduetoMelissaFearonandCarolineFlanaganatSpringerPublishingCompany
whoconvincedustogoaheadwiththisprojectandwereconstantlysupportiveandpatientintheface
of recurring delays and missed deadlines. We thank Paul Drougas from Springer, our editor, for the
helpinimprovingthebookconsiderably.
ThisbookispartiallysupportedbyNSFthroughgrantsIIS-1763365andIIS-1763325.
SomeOtherWords
Broad learning is a fast-growing research area, and few people can have a very deep understanding
about all the detailed materials covered in this book. Although the authors have several years of
explorationexperiencesaboutthefrontierofthisarea,theremaystillexistsomemistakesandtypos
in this book inevitably in writing. We are grateful if the readers can inform the authors about such
mistakestheyfindwhenreadingthebook,whichwillhelpimprovethebookinthecomingeditions.
Tallahassee,FL,USA JiaweiZhang
Chicago,IL,USA PhilipS.Yu
November11,2018
Contents
PartI BackgroundIntroduction
1 BroadLearningIntroduction.................................................. 3
1.1 WhatIsBroadLearning .................................................. 3
1.2 ProblemsandChallengesofBroadLearning................................. 4
1.2.1 Cross-SourceInformationFusion................................... 4
1.2.2 Cross-SourceKnowledgeDiscovery ................................ 6
1.2.3 ChallengesofBroadLearning ..................................... 6
1.3 ComparisonwithOtherLearningTasks ..................................... 7
1.3.1 BroadLearningvs.DeepLearning.................................. 7
1.3.2 BroadLearningvs.EnsembleLearning.............................. 8
1.3.3 BroadLearningvs.TransferLearningvs.Multi-TaskLearning .......... 8
1.3.4 Broad Learning vs. Multi-View, Multi-Source, Multi-Modal,
Multi-DomainLearning........................................... 9
1.4 BookOrganization ...................................................... 10
1.4.1 PartI .......................................................... 10
1.4.2 PartII.......................................................... 11
1.4.3 PartIII ......................................................... 11
1.4.4 PartIV......................................................... 12
1.5 WhoShouldReadThisBook ............................................. 12
1.6 HowtoReadThisBook .................................................. 13
1.6.1 ToReaders ..................................................... 13
1.6.2 ToInstructors ................................................... 13
1.6.3 SupportingMaterials ............................................. 14
1.7 Summary .............................................................. 14
1.8 BibliographyNotes...................................................... 14
1.9 Exercises .............................................................. 15
References................................................................... 15
2 MachineLearningOverview .................................................. 19
2.1 Overview .............................................................. 19
2.2 DataOverview.......................................................... 20
2.2.1 DataTypes ..................................................... 20
2.2.2 DataCharacteristics .............................................. 26
2.2.3 DataPre-processingandTransformation............................. 26
ix
x Contents
2.3 SupervisedLearning:Classification ........................................ 32
2.3.1 ClassificationLearningTaskandSettings ............................ 33
2.3.2 DecisionTree ................................................... 34
2.3.3 SupportVectorMachine .......................................... 40
2.4 SupervisedLearning:Regression .......................................... 46
2.4.1 RegressionLearningTask ......................................... 46
2.4.2 LinearRegression................................................ 46
2.4.3 Lasso .......................................................... 47
2.4.4 Ridge .......................................................... 48
2.5 UnsupervisedLearning:Clustering......................................... 48
2.5.1 ClusteringTask.................................................. 49
2.5.2 K-Means ....................................................... 50
2.5.3 DBSCAN ...................................................... 52
2.5.4 Mixture-of-GaussianSoftClustering................................ 54
2.6 ArtificialNeuralNetworkandDeepLearning................................ 56
2.6.1 ArtificialNeuralNetworkOverview ................................ 57
2.6.2 DeepLearning .................................................. 62
2.7 EvaluationMetrics ...................................................... 66
2.7.1 ClassificationEvaluationMetrics................................... 66
2.7.2 RegressionEvaluationMetrics ..................................... 68
2.7.3 ClusteringEvaluationMetrics...................................... 69
2.8 Summary .............................................................. 71
2.9 BibliographyNotes...................................................... 72
2.10 Exercises .............................................................. 72
References................................................................... 73
3 SocialNetworkOverview ..................................................... 77
3.1 Overview .............................................................. 77
3.2 GraphEssentials ........................................................ 78
3.2.1 GraphRepresentations............................................ 78
3.2.2 ConnectivityinGraphs ........................................... 80
3.3 NetworkMeasures ...................................................... 82
3.3.1 Degree ......................................................... 82
3.3.2 Centrality....................................................... 85
3.3.3 Closeness....................................................... 91
3.3.4 TransitivityandSocialBalance..................................... 96
3.4 NetworkCategories ..................................................... 98
3.4.1 HomogeneousNetwork........................................... 99
3.4.2 HeterogeneousNetwork .......................................... 102
3.4.3 AlignedHeterogeneousNetworks .................................. 106
3.5 MetaPath.............................................................. 111
3.5.1 NetworkSchema ................................................ 111
3.5.2 MetaPathinHeterogeneousSocialNetworks ........................ 111
3.5.3 MetaPathAcrossAlignedHeterogeneousSocialNetworks............. 114
3.5.4 MetaPath-BasedNetworkMeasures ................................ 116
3.6 NetworkModels ........................................................ 117
3.6.1 RandomGraphModel ............................................ 118
3.6.2 PreferentialAttachmentModel..................................... 120
Contents xi
3.7 Summary .............................................................. 121
3.8 BibliographyNotes...................................................... 122
3.9 Exercises .............................................................. 123
References................................................................... 124
PartII InformationFusion:SocialNetworkAlignment
4 SupervisedNetworkAlignment................................................ 129
4.1 Overview .............................................................. 129
4.2 SupervisedNetworkAlignmentProblemDefinition........................... 130
4.3 SupervisedFullNetworkAlignment........................................ 131
4.3.1 FeatureExtractionforAnchorLinks ................................ 132
4.3.2 SupervisedAnchorLinkPredictionModel ........................... 138
4.3.3 StableMatching ................................................. 139
4.4 SupervisedPartialNetworkAlignment ..................................... 142
4.4.1 PartialNetworkAlignmentDescription.............................. 142
4.4.2 Inter-NetworkMetaPathBasedFeatureExtraction.................... 143
4.4.3 Class-ImbalanceClassificationModel............................... 146
4.4.4 GenericStableMatching.......................................... 149
4.5 AnchorLinkInferencewithCardinalityConstraint ........................... 151
4.5.1 LossFunctionforAnchorLinkPrediction ........................... 151
4.5.2 CardinalityConstraintDescription.................................. 153
4.5.3 JointOptimizationFunction ....................................... 154
4.5.4 ProblemandAlgorithmAnalysis ................................... 156
4.5.5 DistributedAlgorithm ............................................ 158
4.6 Summary .............................................................. 160
4.7 BibliographyNotes...................................................... 160
4.8 Exercises .............................................................. 161
References................................................................... 163
5 UnsupervisedNetworkAlignment ............................................. 165
5.1 Overview .............................................................. 165
5.2 HeuristicsBasedUnsupervisedNetworkAlignment .......................... 166
5.2.1 UserNamesBasedNetworkAlignmentHeuristics .................... 166
5.2.2 ProfileBasedNetworkAlignmentHeuristics ......................... 169
5.3 PairwiseHomogeneousNetworkAlignment ................................. 170
5.3.1 HeuristicsBasedNetworkAlignmentModel ......................... 171
5.3.2 IsoRank ........................................................ 173
5.3.3 IsoRankN ...................................................... 175
5.3.4 MatrixInferenceBasedNetworkAlignment.......................... 176
5.4 MultipleHomogeneousNetworkAlignmentwithTransitivityPenalty ........... 177
5.4.1 MultipleNetworkAlignmentProblemDescription .................... 178
5.4.2 UnsupervisedMultipleNetworkAlignment .......................... 179
5.4.3 TransitiveNetworkMatching ...................................... 184
5.5 HeterogeneousNetworkCo-alignment...................................... 186
5.5.1 NetworkCo-alignmentProblemDescription ......................... 187
5.5.2 AnchorLinkCo-inference......................................... 188
5.5.3 NetworkCo-matching ............................................ 195