Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
TaskAssignmentinSpatialCrowdsourcing:ChallengesandApproaches
Hien ToSupervisedbyProf.CyrusShahabiDepartmentofComputerScienceUniversityofSouthernCalifornia
1
ACMSIGSPATIALPhD2016
• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
Outline
2
SpatialCrowdsourcing(SC)
[1] http://www.gartner.com/newsroom/id/2665715
Spatialcrowdsourcing(SC):requiresworkerstophysically traveltothetask'slocationinordertoexecutethetask
Crowdsourcing:outsourcingasetoftaskstoasetofworkers.
SCApplications
Ubiquityofmobileusers
6.5billionmobilesubscriptions,93.5%oftheworldpopulation [1]
Technologyadvancesonmobiles
Smartphone'ssensors.e.g.,video cameras
Networkbandwidth
improvements
From2.5G(up to384Kbps) to3G(upto14.7Mbps)andrecently4G(upto100Mbps)
task
3
SpatialCrowdsourcingApps(a.k.a MobileCrowdsourcing)
MobileCrowdsourcing:TheNextChapter,2016
TheImportanceofMobileCrowdsourcingin2015
There'sanappthatneedsyou:theriseofmobilecrowdsourcing
è TheWiley-AAGInternationalEncyclopediaofGeography4
Outlines• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
5
SpatialCrowdsourcingTaxonomy
6
AssignmentType
Reward-based
Self-incentivised
WorkerSelected
ServerAssigned
PointTask
RegionTask
SpatialTaskType
WorkerType
ComplexTask
SpatialCrowdsourcing
PushPull
ReportType ExplicitReport
ImplicitReport
CrowdsensingCrowdsourcing
Privacy QualityControlIncentives
Reporting
Tasking
TravelcostQuality(quantity)ofassignedtasks
Communicationcost
Totalrewards
ChallengesMultipleconsiderations
Multipleobjectives
ResearchChallengesinSpatialCrowdsourcing
• Asghari etal.SIGSPATIAL2016• Bessai andCharoy ISCRAM‘16• HassanandCurryESA’16• Zhangetal.TVT’16• Gaoetal.WAIM’16• Chengetal.TKDE‘16• Tongetal.VLDB‘16• Liuetal.DASFAA‘16• Huetal.ICDE’16• Tongetal.ICDE’16• Zhangetal.WCMC‘16• Liuetal.UbiComp ‘16• Guo etal.THMS’16• Toetal.PerCom ’16
7
• Toetal.TSAS’15• Alfarrarjeh etal.MDM’15• Fonteles etal.MoMM ‘15• HassanandCurry.SIGSPATIAL’15• Xiaoetal.INFOCOM’15• Xiong etal.PerCom ‘15• Pournajaf etal.ICCS‘14• HassanandCurry.UCI’14• Heetal.INFOCOM’14• Fonteles etal.SIGSPATIAL‘14• Zhangetal.UbiComp ’14• Dangetal.iiWAS ‘13• Kazemi andShahabi.SIGSPATIAL‘12
Taskassignment(orworkerselection)Processofidentifyingwhichtasksshouldbeassignedtowhichworkers
ResearchChallengesinSpatialCrowdsourcing
8
Privacy-preservingSCProtectlocationsofworkersandtasks
• Toetal.TMC’16• Boutsis andKalogeraki.UbiComp‘16• Zhangetal.CN‘16• Zhangetal.ATIS‘15• Shenetal.GLOBECOM‘15• Gongetal.IoT’15• Gongetal.TETC’15• Huetal.APWeb ’15• Pournajaf etal.MDM’14,
SIGSPATIAL’15• Toetal.VLDB’14,ICDE’15• Boutsis andKalogeraki PerCom ‘13• Vuetal.INFOCOM‘12• Kazemi andShahabi SIGKDD’11
TaskschedulingWorkers’pathplanningtoperformtasks
• Wangetal.2016• Fonteles etal.JLBS‘16• Dengetal.GeoInformatica ’16• Mrazovic etal.ICDMW‘15• Chenetal.IJCAI‘15• Chenetal.AAMAS ‘15• Hadano etal.HCOMP’15• Dengetal.SIGSPATIAL’15• Chenetal.HCOMP‘14• DengandShahabi. SIGSPATIAL’13• Sadilek etal.AAAI’2013
ResearchChallengesinSpatialCrowdsourcingTrustandqualityConsiderqualityofreportdataortrustworthinessofworkers
• Liuetal.Sensor’16• Zhangetal.TETC’16• Miaoetal.DSS’16• Fanetal.SOSE’15• Anetal.HPCC’15• Kangetal.MASS’15• Cheng etal.VLDB’15• Zhaoetal.MDM’15• Wangetal.UbiComp ’15• Songetal.TVT’14• Boutsis etal.ICDCS’14• Fengetal.INFOCOM’14• Kazemi etal.SIGSPATIAL’13
9
IncentivemechanismIncentivizeworkerstoperformtasks
• Zhangetal.TVT’16• Kandappu etal.CSCW‘16• Kandappu etal.UbiComp ‘16• Micholia etal.IJHCS’16• Toetal.GeoRich ’16• LiandCaoTMC’16• Guo etal.IJMHCI‘16• Thebault-Spieker etal.CSCW’15• Jin etal.MobiHoc ’15• Fanetal.ICTAI’15• Shah-Mansouri etal.ICC’15• Teodoroetal.CSCW’14• Rula etal.HotMobile ‘14• Musthag etal.CHI’13• Heimerl etal.CHI’12• Jainmes etal.PerCom ‘12• Yangetal.MobiCom ‘12• LeeandHohPMC‘10• Altetal.NordiCHI ‘10
ResearchChallengesinSpatialCrowdsourcing
Applications• KonomiandSasao Urb-IoT ‘16• Jaiman etal.UbiComp/ISWC’16• FanandTsengMOBIS‘15• KonomiandSasao UbiComp/ISWC’15• Harburg etal.CHI’15• Chenetal.SenSys ‘15• KimCHI‘15• Aubry etal.CROWDSENSING’14• Chenetal.VLDB’14• Kimetal.MMSys’14• Benouaret etal.IEEEIC’13• Coric andGruteser DCOSS’13• Koukoumidis etal.MobiSys ‘11• Goodchild andGlennon IJDE‘10
10
GenericframeworksDiscusscomponents,architecture,programmingframeworkofSCapps
• Toetal.CROWDBENCH’16• Fonteles etal.RCIS‘16• Pengetal.ASE’16• Kucherbaev etal.SIGCHI’16• Sakamotoetal.COMPSAC’16• Fernandoetal.MOBIQUITOUS’13• Tamilin etal.UbiComp ‘12• Raetal.MobiSys ‘12• Yanetal.SenSys ‘09
Relatedsurveys• Pournajaf etal.SIGMODRec.‘16• Guo etal.CompSurvey‘15• ZhaoandHan2016• ChristinJSS’15
• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
Outlines
11
RisksofLocationLeaks
SecurityFlawInGayDatingAppGrindrRevealsPreciseLocationOf90%OfUsers
Grindr’sLocationSecurityUpdate“AnyuserwhoconnectstoGrindrinthesecountrieswillhavetheirdistancehidden automaticallybydefault,whichincludeRussia, Egypt,SaudiArabia,Nigeria,Liberia,etc.”
CouldusinggaydatingappGrindrgetyouarrestedinEgypt?
StalkersExploitCellphoneGPS“25,000adultsintheU.S.arevictimsofGPSstalkingannually, includingbycellphone.”byU.S.JusticeDepartment,Aug2010.
TheHiddenPrivacyThreatof…FlashlightApps?
“~50%topiPhones andAndroidsmartphonesdisclosed auser’s locationtothirdpartieswithoutuserconsent.”byWallStreetJournal,Dec2010.
Apple,GoogleCollectUserData
12
PrivacyImplicationofLocationRelease
“InferenceAttacksonlocationtracks”[JohnKrumm,PERVASIVE'07]• GPSdatafrom172people• Findhomelocationswithmedianerrorof60meters
“Understandingindividualhumanmobilitypatterns”[Baraba´sietal.,Nature’08]
• 100,000mobilephoneusers,6months• Usermobilityà spatialprobabilitydistribution
“UniqueintheCrowd:” [Montjoye etal.,Nature’13]• 1.5millionindividuals,15months• 4 locationsuniquelyidentify95% individuals
13
PrivacyImplicationofLocationRelease
• Locationleakssensitiveinformation• Politicalandreligiousview• Individual’shealthstatus• Alternativelifestyles• Affiliations
• Attacksbasedonlocations• Physicalsurveillance• Stalking• Identitytheft
14
• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
Outlines
15
16
TaskRabbit:ACaseStudy[Wiki]
isanonlineandmobilemarketplacethatmatchesfreelancelaborwithlocaldemand,allowingconsumerstofindimmediatehelpwitheverydaytasks,includingcleaning,moving,delivery andhandymanwork
Leakoftasklocation duringtasking
A deliverytask
Cloakedtasklocations
17
TaskRabbit:ACaseStudyLeakofworkerlocationtaskingandreporting
Cloakedtasklocations Taskstatus
18
TaskRabbit:ACaseStudyLearnarequester’shomebytrackinghisrequests• Hypothesis:cloakedtaskregionscoveractualtasklocation• Validatethehypothesis:aknownlocationofanothertaskis
actuallyintheoverlappedarea
Triangulationattack
• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
Outlines
19
CurrentState-of-the-artCategory Paper Description
Tasking&Reporting
ShinPMC‘11 Protectingworkers’ identityandlocation (IPaddress)during task download(useTor)andreportsubmission(useanonymity service)
Reporting Boutsis PerCom ‘13• ZhangCN‘16
Workersexchange spatialtaskreportsbeforeuploadingtoSC-serversothattheservercannottellwhichworkersuploadwhichreports
Tasking Kazemi SIGKDDEx’11• VuINFOCOM‘12
SC-serverdistributesspatialtaskstoworkerssuchthateachworkerisassignedasubsetoftasksthatareclosertohimself thantoanyotherworker
ToVLDB’14• GongTETC’15• ZhangATIS‘15• ToTMC’16
SC-serverassignsaspatialtasktoasetofworkers• usedifferentialprivacytohideworkers’ locationwith
lowtravelcostandlowcommunicationoverheadwithoutcompromising assignmentsuccessrate
Pournajaf MDM’14• Hu APWeb ’15
SC-servermatches asetofspatialtaskstoasetofworkersgiventheircloaked regions• maximizetaskassignmentwhilesatisfyingtravel
budgetconstraintofeachworkerShenGLOBECOM‘15
SC-serverassignsaspatialtasktothe nearest workersontheencrypted domain
SpatialCrowdsourcingTaxonomy
21
AssignmentType
Reward-based
Self-incentivised
WorkerSelected
ServerAssigned
PointTask
RegionTask
SpatialTaskType
WorkerType
ComplexTask
SpatialCrowdsourcing
PushPull
ReportType ExplicitReport
ImplicitReport
CrowdsensingCrowdsourcing
Privacy QualityControlIncentives
Reporting
Tasking
TravelcostQuality(quantity)ofassignedtasks
Communicationcost
Totalrewards
ChallengesMultipleconsiderations
Multipleobjectives
𝑡"
SC-server(Crowdsourcing
Company)t1t2t7
Requesters
SC-ServerPerformsTaskAssignment[*]
• SC-serverchoosesbestworkersfortaskbasedontask-workerproximity
• SC-serverknowsworkers’locations
𝑡#𝑡$
𝑡%
𝑡&
𝑡'
𝑡()
22[*] Kazemi and Shahabi. GeoCrowd: Enabling Query Answering with Spatial Crowdsourcing. SIGSPATIAL 2012
WhyDifferentialPrivacy [*]?
Privacyguaranteesü Protectionforindividualsü Protection againstbackgroundknowledgeü Strongtheoreticalguaranteesonprivacyofreleaseddata
23
Differentialprivacysupportsvariousaggregationqueries,e.g.,count,min,max,mean,etc.
PrivatizedResults
PrivatizedAnalysisDatabase
𝑅𝐷 𝑄(𝐷) = 𝑅
[*] Dwork, Nissim, McSherry, Smith, Calibrating Noise to Sensitivity in Private Data Analysis 2006
AdoptionofDifferentialPrivacy
“iOS10addsSiriintelligenceintoQuickType andPhotos,…,andopensupSiri,Maps,PhoneandMessagestodevelopers—whileincreasingsecurityand privacy withpowerfultechnologieslikeDifferential Privacy.”CraigFederighi, Apple’sseniorvicepresidentofSoftwareEngineering,June13,2016.
“RAPPOR enables learning statistics about the behavior of users’ softwarewhile guaranteeing client privacy. The guarantees of differential privacy,which are widely accepted as being the strongest form of privacy, ….RAPPOR introduces a practical method to achieve those guarantees.” ÚlfarErlingsson, Google’s Tech Lead Manager, Security Research
AppleandGooglehaveadaptedDPtodiscoverusagepatternsfromalargenumberofusers.
DifferentiallyPrivatePublicationofLocationEntropyThursday,November3.8:00- 9:00AM(Session11)
DifferentialPrivacyinUse
25
OutputofQ:5+symmetricrandomLaplacenoise
Inspatialcrowdsourcing,𝑄 = “Howmanypeopleinaparticulararea?”
UsingDifferentialPrivacyinTaskAssignmentphaseofSpatialCrowdsourcingtohidethelocationofworkerswithlowcost andlowoverheadwithoutcompromisingperformance.
ProblemStatement
• Lowcost:travelcost• Lowoverhead: communicationoverheadrequiredtosendtaskrequeststoworkers
• Performance:#ofperformedtasks
26
Privacy-PreservingTaskAssignmentPerformtaskassignmentinDP-compliantmanner• UseworkerdatasetPrivateSpatialDecomposition(PSD)
Challenges:1. DPintroduces“fake”workers- mayleadto
inefficientassignment• Assigneachtasktoallworkersisnotscalable
• Ensuretaskcompletion,butkeepoverheadlow,e.g.,minimizethenumberofworkersthatneedtobenotified
2. Communicationmodelsignificantlycomplicated• EstablishingdirectcommunicationchannelsviolatesDP!
• Instead,disseminatetaskrequestswithingeocast region[*]
[*] Navas and Imielinski. Geocast: geographic addressing and routing, MOBICOM’97 27
0 3 3 2
4 7 6 3
4 5 2 3 1
2 2 23 6
3 3 2
1.CSPaggregatelocations
3.SC-serverreceivestaskrequest
4.SC-serverqueriesPSDtofindgeocast region(GR)thatenclosessufficient workers
5.SC-servergeocast thetasktoallworkerswithinGR
2.CSPreleasesPSD,accordingtoDP
SystemArchitecture
Aggregateslocations
CellServiceProvider
SC-serverSanitizeslocations
Consent
Geocast[*]
PSD
Workers
[*] Navas and Imielinski. Geocast: geographic addressing and routing, MOBICOM’9728
PerformanceMetricsandResults
AssignmentSuccessRate(ASR)• Ratiooftasksacceptedbyaworker
WorkerTravelDistance(WTD)• Captures travelcostorassignmentquality
AverageNotifiedWorkers (ANW)• Capturesoverhead(communicationrequiredto
geocast taskrequestsorcomputationofassignmentalgorithm)
29150% increasecomparedtonon-privatecase
20% increasecompared tonon-privatecase
Similarperformance compared tonon-privatecase
PublishingResults
[Toetal.TMC2016] [2]
• Protectingworkerlocationsacrossmultipletimestamps
[Toetal.VLDB2014]• Differentially-privateframeworkfortaskassignment
[Toetl al.ICDE2015] [1]• Toolboxforturningspatialcrowdsourcingparameters
[1] To et al. PrivGeoCrowd: A Toolbox for Studying Private Spatial Crowdsourcing, ICDE 2015[2] To et al. Differentially Private Location Protection for Worker Datasets in Spatial Crowdsourcing, TMC 201630
• Introduction• Ataxonomyandresearchchallenges• Privacy-preservingspatialcrowdsourcing– Risksoflocationleaks– Privacythreatsduringtaskingandreporting• TaskRabbit:acasestudy
– Privacycountermeasures:thestate-of-the-art• Conclusionandfutureworks
Outlines
31
• Taxonomyofspatialcrowdsourcing• Moststudiesfocusontaskassignmentproblem– Multipleobjectives:number(quality)ofassignedtasks,totalrewards,travelcost,communicationoverhead,etc.
– Multipleconsiderations:privacy,trust,quality,incentive
• Privacyissue– Privacyrisksoccurtoworkers andrequestersduringtaskingandreportingphases
– Varioustechniqueshavebeenused:pseudonymity,cloaking,perturbation,encryption,exchange-based
• Taskassignmentwithrigorousprivacyprotection– Thecostofprivacyispractical– Butnonesolvesprivacyrisksforbothworkersandtasks
Conclusionsandfutureworks
32