1 Understanding the Role of Places and Activities on Mobile Phone Interaction and Usage Pa￿erns ABHINAVMEHROTRA,UniversityCollegeLondon,UK SANDRINER.MU¨LLER,UniversityofCambridge,UK GABRIELLAM.HARARI,StanfordUniversity,USA SAMUELD.GOSLING,UniversityofTexasatAustin,USAandUniversityofMelbourne,Australia CECILIAMASCOLO,UniversityofCambridgeand￿eAlanTuringInstitute,UK MIRCOMUSOLESI,UniversityCollegeLondonand￿eAlanTuringInstitute,UK P.JASONRENTFROW,UniversityofCambridge,UK Userinteractionpa￿ernswithmobileappsandnoti￿cationsaregenerallycomplexduetothemanyfactorsinvolved.However adeepunderstandingofwhatin￿uencesthemcanleadtomoreacceptableapplicationsthatareabletodeliverinformationat therighttime.Inthispaper,wepresentforthe￿rsttimeanin-depthanalysisofinteractionbehaviorwithnoti￿cationsin relationtothelocationandactivityofusers.Weconductedanin-situstudyforaperiodoftwoweekstocollectmorethan 36,000noti￿cations,17,000instancesofapplicationusage,77,000locationsamples,and487daysofdailyactivityentriesfrom 26studentsataUKuniversity. Ourresultsshowthatusers’a￿entiontowardsnewnoti￿cationsandwillingnesstoacceptthemarestronglylinkedtothe locationtheyareinandinminorparttotheircurrentactivity.Weconsiderbothusers’receptivityanda￿entiveness,andwe showthatdi￿erentresponsebehaviorsareassociatedtodi￿erentlocations.￿ese￿ndingsarefundamentalfromadesign perspectivesincetheyallowustounderstandhowcertaintypesofplacesarelinkedtospeci￿ctypesofinteractionbehavior. ￿isinformationcanbeusedasabasisforthedevelopmentofnovelintelligentmobileapplicationsandservices. CCSConcepts:•Human-centeredcomputing HCIdesignandevaluationmethods;Empiricalstudiesinubiquitousand ! mobilecomputing; AdditionalKeyWordsandPhrases:MobileSensing,Noti￿cations,ApplicationUsage,Context-awareComputing. ACMReferenceformat: AbhinavMehrotra,SandrineR.Mu¨ller,GabriellaM.Harari,SamuelD.Gosling,CeciliaMascolo,MircoMusolesi,andP.Jason Rentfrow.2017.UnderstandingtheRoleofPlacesandActivitiesonMobilePhoneInteractionandUsagePa￿erns.PACM Interact.Mob.WearableUbiquitousTechnol.1,1,Article1(July2017),22pages. DOI:10.1145/nnnnnnn.nnnnnnn 1 INTRODUCTION Mobilephonestodayhavebecomeanindispensablepartofourdailylives.Farfrombeingsimplecallinginstru- ments,theyarenowadvancedcomputingplatformswithalways-onconnectivity,high-speeddataprocessingand advancedsensing[21].￿esea￿ordanceshaveopenedthepossibilityofimplementingnovelcontext-awareand personalizedapplicationsthatareabletoassistusinavarietyofday-to-daysituations.Atthesametime,they Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthat copiesarenotmadeordistributedforpro￿torcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthe￿rst page.CopyrightsforcomponentsofthisworkownedbyothersthanACMmustbehonored.Abstractingwithcreditispermi￿ed.Tocopy otherwise,orrepublish,topostonserversortoredistributetolists,requirespriorspeci￿cpermissionand/orafee.Requestpermissionsfrom [email protected]. ©2017ACM. 2474-9567/2017/7-ART1$15.00 DOI:10.1145/nnnnnnn.nnnnnnn PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. 1:2 • Mehrotraetal. (a) (b) (c) (d) Fig.1. TheMyLifeLoggerapplication. representauniqueplatformforreal-timedeliveryofinformationaboutavarietyofeventsrangingfromemails toupdatesononlinesocialnetworks,fromadvertisementstopositivebehaviorchangeinterventions[5,20,31]. ￿eadventofmobilesensinghasprovidedagreatopportunityforresearchersandpractitionerstoinvestigate users’mobileinteractionbehavior.Mobileappsarenowcapableofrecordinginformationaboutusers’interactions withmobilephonesandthesurroundingcontext(e.g.,location,activity,audioenvironment,andcollocationwith otherdevices).￿ankstotheavailabilityofthisinformation,researchershavebeenabletoconductseveralstudies onavarietyofaspectsofhuman-smartphoneinteraction.Forexample,studieshaveinvestigatedthediversityin appusagebehaviorofindividuals[14],thecharacterizationofmobileusageinruralandurbansocieties[13] orindi￿erentsocio-economicgroups[32]. Othershavefocussedonthein￿uenceofpersonality[10,11],the associationbetweensocialcontextandappusagepa￿erns[26],andusers’motivationsforusingdi￿erenttypes ofapps[18].Moreover,somestudieshavealsofocusedonunderstandingusers’appusagepa￿ernsforpredicting theirfutureinteractionwithapps[27,38]. Anotherparticularaspectthathasa￿ractedthea￿entionofresearchersgivenitspracticalimportanceisthe characterizationofthereactionofuserstonoti￿cationsandthedesignofmobilenoti￿cationmanagementsystems. Forexample,somestudieshaveinvestigatedthefactorsthatin￿uenceusers’a￿entivenessandreceptivityto noti￿cations[24,30]andhowthesearein￿uencedbycontext[28],content[23],andthecomplexityofanongoing task[24].Otherprojectshaveaimedatanticipatingusers’a￿entiveness[30]andreceptivity[22]tonoti￿cations bylearningtheirbehavioralpa￿erns. However,existingworkhasnotconsideredtheimpactoftheexternalfactors,suchasthetypeoflocations usersareinandtheactivitiestheyarecurrentlycarryingout,ontheirinteractionwithnoti￿cationsandapp usage behavior. A deep understanding of these factors would enable us to improve users’ experience and the e￿ectiveness of noti￿cations as well as applications (e.g., marketing and positive behavior intervention applications). Keychallengesforsuchastudyincludedatacollectionatavery￿negranularity,whichmight requirefrequentinputsfromusers,andtheextractionofhigh-levelinformationfromrawsensordatawiththe goalofassigningsemanticstoit. PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. UnderstandingtheRoleofPlacesandActivitiesonMobilePhoneInteractionandUsagePa￿erns • 1:3 Tobridgethisgap,inthispaperwepresentthe￿rstin-situstudyoftheimpactoflocationandactivitieson users’interactionwithmobilenoti￿cationsandapplications. Overaperiodoftwoweeks,wecollectedmore than36,000noti￿cations,17,000instancesofapplicationusage,77,000locationsamples,and487daysofdaily activityentries(i.e.,2984activityinstances)from26studentsataUKuniversity.Usingthisdata,weinvestigate users’interactionwithmobilenoti￿cationsandappswhentheyareperformingdi￿erentactivities,whenthey visitdi￿erenttypesoflocations,andlocationswithdi￿erentcharacteristicssuchasbeingboringvs.exciting,sad vs.happy,inactivevs.busy,lazyvs.productive,distressingvs.relaxing,andnaturalvs.urban. ￿ekeycontributionsand￿ndingsofthisworkcanbesummarizedasfollows: Wediscussanin-depthstudyoftherelationshipsbetweenusers’activitiesandinteractionwithnoti￿ca- • tionsandapps. Wediscussaquantitativeevaluationofusers’interactionwithnoti￿cationsandappswhentheyareat • di￿erenttypesoflocations;thisisimportantfromadesignperspective,sinceitallowsustounderstand howcertaintypesofplacesarelinkedtospeci￿ctypesofinteractionbehaviorasabasisofthedevelopment ofintelligentapplications. Wepresentanextensiveinvestigationoftheimpactoflocationcharacteristicsonusers’a￿entiveness • andreceptivitytonoti￿cationsandappusagepa￿ern.Weconsiderdi￿erentcharacterizationofplaces (e.g.,urban,productiveandsoon)inrelationtousers’interactionwithnoti￿cations.Again,besidesthe inherentintellectualinterest,the￿ndingsmightbeusedinthedesignofpersonalizedapplicationsbased onlyontheknowledgeofthecurrentlocation. ￿roughouranalysisweuncovervariousnewinsightsaboutthephoneusageandnoti￿cationinteraction behaviorsofusersandalso,quiteimportantly,con￿rmsome￿ndingsofpreviousstudies.Morespeci￿cally,the mainnovel￿ndingsofouranalysisare: Participantsweremorereceptivetonoti￿cationswhiletheywereexercisinganddoingroutinetasks. • Overallcommunicationappswereusedthemost,exceptwhileparticipantsweregoingtosleep,whichis • whentheymostlyusedlifestyleapps. Participantsweremorereceptivetonoti￿cationswhentheywereatcollege,inlibraries,onstreets,and • theywereleasta￿entivewhilebeingatreligiousinstitutions. ￿eappusagewashighestwhileparticipantswereatcollegeorinlibraries. • Participantsusedmostlymusicandreadingappswhilewaitingatbusstopsandtrainstations. • Participantsweremorea￿entivetonoti￿cationsatproductiveplacescomparedtolazyplaces. • Participantswerelessreceptivetonoti￿cationsatnaturalplacescomparedtourbanplaces. • Whileatlazy,distressingandnaturalplacesparticipantsusedtheirphoneslesscomparedtoproductive, • relaxingandurbanplacesrespectively. ￿isstudyhasledtomanyinterestinginsightsthatarediscussedindetailinthefollowingsections.Firstof all,peoplepaylessa￿entiontonoti￿cationswhentheyarepreparingtogotobed(ortheyareinbedbefore sleeping)andwhileexercising.Peopleacceptmostnoti￿cationsthataredeliveredwhiletheyaredoingexercise androutinetasks.Whenpeopleareonstreets,orincollege,universityandresidentialareas,theynotonlypay morea￿entiontowardsnoti￿cationsbutalsoacceptmostofthem.Also,peoplearemorea￿entivetonoti￿cations atplacesthatarecharacterizedas“productive”.￿eyacceptmorenoti￿cationsatproductiveandurbanplaces. Furthermore,overallphoneusagedurationaswellasusageofspeci￿cappsvarywhenpeoplevisitspeci￿ctypes ofplacesandwhileperformingspeci￿cactivities. Overallappusageishighestwhilepeoplearerelaxing,at collegeorinlibraries,whereasitislowestwhilepeoplearedoingexercise,routinetasksorgoingtosleep,and whentheyareatgyms,religiousinstitutionsorinparkingplaces. Webelievethatthepotentialapplicationsofthisworkaremany. Firstofallthe￿ndingsofthispapercan be used as a basis for the development of predictive applications that rely on the analysis of users’ current PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. 1:4 • Mehrotraetal. locationsandnotonlyontheirpastbehavioralpa￿erns.Morespeci￿cally,thisisparticularlyimportantinthe bootstrappingphaseofintelligentapplicationsthatarebasedonlearningalgorithmsthatrequirealargehistory ofpastinteractionswiththephonesinordertomakeaccuratepredictionaboutusers’behavior.Examplesinclude noti￿cationmanagementsystems,andpre-cachingandpre-launchingmechanismsformobileapplications. 2 RELATEDWORK Inthissectionwereviewtherelatedworkintwokeyareas,namelythestudiesaboutthecharacterizationof users’interactionwithnoti￿cationsandthoseaboutusers’appusagebehavior. 2.1 UnderstandingUsers’InteractionwithNotifications Inrecentyearstheareaofmobileinterruptibilityhasreceivedincreasinga￿ention.Previousstudieshaveexplored variousaspectsofusers’interactionwithmobilenoti￿cations[4,6,7,12,24,29,35].Inparticular,in[35]Sahami etal.showthatusersdealwitharound60noti￿cationsperday,andmostoftheseareviewedwithinafewminutes ofarrival.Additionally,bycollectingthesubjectivefeedbackfrommobileusers,theauthorsdemonstratethat usersassigndi￿erentimportancetonoti￿cationstriggeredbyapplicationfromdi￿erentcategories.Atthesame time,Pielotetal.[29]showthatpersonalcommunicationnoti￿cationsarerespondedtoquickestbecauseofsocial pressureandtheexchangeoftimecriticalinformationthroughcommunicationapplications(i.e.,Whatsapp). Ontheotherhand, asIqbaletal. suggestin[19], usersarewillingtotoleratesomedisruptioninreturnfor receivingnoti￿cationsthatcontainvaluableinformation.Similarly,in[24]Mehrotraetal.showthatnoti￿cations containingimportantorusefulcontentareo￿enaccepteddespitethedisruptioncausedbythem. Moreover,otherstudieshavealsoinvestigatedhowusers’a￿entivenessandreceptivitytonoti￿cationsare in￿uencedbytheircontext[28,30]andcontent[16,22,23]. In[28],theauthorsshowthatthea￿entiveness ofuserscanbedeterminedbycontextualfactorsincludingactivity,locationandtimeofday. ￿eyproposea mechanismthatreliesonthesecontextmodalitiestopredictopportunemomentsfordeliveringnoti￿cations. In[30]Pielotetal.proposeamodelthatcanpredictwhetherauserwillviewanoti￿cationwithinafewminutes withaprecisionofapproximately81%. Ontheotherhand,in[16]theauthorsshowthatusers’receptivityis in￿uencedbytheirgeneralinterestinthenoti￿cationcontent,entertainmentvalueperceivedinitandaction required by it, but not the time of delivery. In [23] Mehrotra et al. suggest to use contextual information, sender-recipientrelationshipandapplicationcategorythattriggeredthenoti￿cationfordeterminingtheuser’s interruptibility.Inanotherstudy[22],theauthorsdemonstratethatusers’receptivitytonoti￿cationsisin￿uenced bytheirlocationandthecontentofnoti￿cationsdelivered.￿eauthorsproposeasystemthatreliesonmachine learningalgorithmsfortheautomaticextractionofrulesthatre￿ectuser’spreferencesforreceivingnoti￿cations indi￿erentsituations. 2.2 UnderstandingUsers’AppUsageBehavior Previous studies have investigated the association between users’ app usage behavior and various socio- economic[13,14,32]andpsychological[10,11,26]factors.Othershavefocussedonusers’motivationsforusing di￿erenttypesofapps[8,18]. Morespeci￿cally, in[32], Rahmatietal. presentastudythatinvestigateshowuserswithacertainsocio- economicstatusinstallanduseapps.￿eir￿ndingscon￿rmthein￿uenceofsocio-economicstatusonphone usage. In[11], theauthorsshowthatthereisasigni￿cantassociationbetweenusers’personalitytraitsand phoneusage.Furthermore,quiteinterestingly,in[18]Hinikeretal.provideevidencethatusers’motivationsfor engagingwithtechnologycanbedividedintoinstrumentalandritualistic. In[8],Bohmeretal. presentalarge-scalestudywiththegoalofunderstandingusers’appusagepa￿erns basedontheircontext.￿e￿ndingsofthisstudydemonstratethatusersspendaroundanhoureverydayusing theirphones,buttheiraveragesessionusinganapplicationlastsforaminute.Overall,communicationappsget PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. UnderstandingtheRoleofPlacesandActivitiesonMobilePhoneInteractionandUsagePa￿erns • 1:5 Group Feature Description Arrivaltime Timeatwhichanoti￿cationarrivesinthenoti￿cationbar. Removaltime Timeatwhichanoti￿cationisremovedfromthenoti￿cationbar. Noti￿cation Senderapplication Nameandpackageoftheapplicationthattriggersthenoti￿cation. AppName Nameoftheapplication. ApplicationUsage LaunchTime Timeatwhichtheapplicationislaunchedandappearedintheforeground. BackgroundTime Timeatwhichtheapplicationuseisendedanditismovedfromforeground tobackground. Lock/unlockevent Timeatwhichthephonewaslockedandunlocked. PhoneInteraction Screeninteraction Typeofinteraction(i.e.,singleclick,longclickandscroll),timeandthename offoregroundapplications(includinghomescreen)withwhichtheinteraction happened. Location Geo-locationoftheplacesvisited. Context DailyActivity Typeandtimedurationofdi￿erentactivitiesperformedinaday.￿eseactiv- itiesincludesleep,eat,work,physicalexercise,socialactivityandrelaxation. Table1. Descriptionoffeaturesfromthedataset. usedmost,exceptwhenusersaretraveling,inwhichcasetheyaremorelikelytousemultimediaapps.In[15], Ferreiraet.al.conductedastudyshowingthatappusagebehaviorofusersisstronglyin￿uencedbytheirsocial andspatialcontext. Also,Xuetal.[37]exploitthenetworktra￿cfromapps(basedonHTTPsignatures)to demonstratethatappusageisin￿uencedbyspatialandtemporalfactorsincludinggeographicalareasandtime oftheday.￿eir￿ndingsalsoshowthatcertainappshaveanonnegligiblelikelihoodofco-occurrence. An open question in this area remains the impact of locations and activities on users’ interaction with noti￿cations. ￿epresentstudyaimstobridgethisgapbyinvestigatingthee￿ectsofthesefactorsonusers’ a￿entivenessandreceptivitytonoti￿cationsaswellasontheirappusagebehavior. 3 METHODOLOGY Inthissectionwepresentourapproachforinvestigatingthein￿uenceofdailyactivities,typeandcharacteristics ofvisitedlocationsonusers’appusage,andnoti￿cationinteractionbehavior. 3.1 LifeLoggerApp Giventheaimsoftheproposedinvestigation,wedesignedandcarriedoutanin-the-wild study[34]tocollect users’data.Morespeci￿cally,wedevelopedanAndroidappcalledMyLifeLogger (showninFigure1).￿eapp performs continuous sensing in the background to log users’ interaction with noti￿cations, app usage, and context.Table1providesadescriptionofthefeaturescapturedbyMyLifeLogger. ￿eappreliesonAndroid’sNoti￿cationListenerService[1]andUsageStatsManager[2]totracenoti￿cations andapplicationusage. Moreover, theappallowsuserstoprovidetheirdailyactivityschedules, forwhicha remindernoti￿cationistriggeredeverynightat9pm(localtime).AsshowninFigure1(b),usersweregivenalist ofsixpossibledailyactivities: Eat:timeperiodwhenauserishavingfood. • Sleep:timeperiodforwhichauserslept.1 • Work:timeperiodwhenauserisengagedinanactivityinvolvingmentalorphysicale￿ort.Sinceour • participantsarestudents,thisactivitywouldmostlyconsistoforberelatedtostudying. 1Itisworthnotingthatparticipantsdidnotreceiveanystandardde￿nitionoftheseactivitiesbeforethestudy.￿erefore,forexample,some participantsmighthaveinterpretedsleepingasbeinginbed. PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. 1:6 • Mehrotraetal. Exercise:timeperiodwhenauserisperforminghealthand￿tnessactivities. • Social:timeperiodwhenauserissocializingwithothers. • Relaxation:timeperiodduringwhichauserisbeingfreefromtensionandanxiety. • Itisworthnotingthatparticipantswereabletoselectonlyoneactivityforaspeci￿ctimeinterval.Moreover, theywereallowedtoenterotheractivitiesbyselectingtheother option,throughwhichtheycouldinputan activitynameasfreetext. Mostactivitiesregisteredthroughtheother activityoptionwererelatedtoroutine taskssuchaslaundry,cooking,ge￿ingready,packing,supermarketandsoon.￿erefore,wecreatedanother categoryforchoresandmappedtheroutinetasksenteredthroughtheother activityoptiontothisnewactivity. ￿eMyLifeLoggerappalsocollectedadditionaldataaboutothercontextualfeatures(suchasmovement,call andSMSlogs)aswellasmood-relatedquestionnaires. However,wedonotdiscussthoseaspectsofthedata becausetheyarenotusedfortheanalysispresentedinthispaper. 3.2 RecruitmentoftheParticipants MyLifeLoggerwaspublishedonGooglePlayStore2andadvertisedto￿rst-yearundergraduatestudentsataUK University.Itwasinstalledby28studentsand26studentscompletedthestudybykeepingtheapprunningon theirphoneforaminimumoftwoweeks.￿eseparticipantscomefrombothsexes(16malesand10females), withtheagespanbetween18and27years(mean =19.46andstandardde�iation =2.18). ￿estudentswere enrolledin15di￿erentcoursesand27%(n=7)werenon-British.Allparticipantswhocompletedthestudywere given£25. 3.3 EnsuringPrivacyCompliance InordertoallowtheMyLifeLoggerapplicationtomonitornoti￿cationsandappusage,theuserhastogiveexplicit permissionasrequiredbytheAndroidoperatingsystem.Moreover,theapplicationalsoshowsaconsentform detailingtheinformationthatiscollected.￿isensuresthattheusergoesthroughatwo-leveluseragreement andiscompletelyawareofthetypeofinformationcapturedbytheapplication. Itisworthnotingthatwereceivedethicalapprovalforthestudy,includingallproceduresandmaterials,from thePsychologyResearchEthicsCommi￿eeattheUniversityofCambridge. 3.4 Dataset Weanalyzedthedataof26userswhoparticipatedforaminimumperiodoftwoweeks.￿edatasetcorresponding totheseusersincludes36,106noti￿cations,17,680instancesofapplicationusage,77,306locationsamples,and 487daysofdailyactivityentries(i.e.,2,984activities). 3.4.1 CharacterizingLocations. Inordertoperformouranalysisontheimpactoflocationonnoti￿cation responseandapplicationusage,wecannotjustrelyonsensor(i.e.,GPS)dataasitonlyprovidesthecoordinates ofaplaceratherthanitstypeandcharacteristics.￿erefore,wemanuallycategorizedthelocationsbyclustering themintosigni￿cantplaces,andthencharacterizedthesigni￿cantplacesbyhavingcodersrateeachplaceon severaldimensions(e.g.,thedegreetowhichaplaceisinactivevsbusy). IdentifyingSigni￿cantPlaces.Firstofall,wediscardthelocationsampleswithmorethan50metersaccuracy sothattheestimatedlocationclustersareofbe￿erquality.Wethen￿ndthelocationsamplesthatwerecollected whileusersweremovingandwealsodiscardthem.Inordertoinfersuchlocationpoints,wecomputethespeed oftheuserbyusingthedistanceandthetimebetweenthelastandthecurrentlocationpoints.Ifthespeedis 2h￿ps://play.google.com/store/apps/details?id=com.nsds.mystudentlife PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. UnderstandingtheRoleofPlacesandActivitiesonMobilePhoneInteractionandUsagePa￿erns • 1:7 lessthanacertainthreshold(i.e.,5kmperhour)weconsiderthatlocationreadingwascollectedwhentheuser wasnotmoving. Now,weusethelocationclusteringapproachpresentedin[36]forgroupingthe￿lteredlocationsamples.We iterateoveralllocationsamplesandforeachlocationpointwecreateanewclusteronlyifthedistanceofthis locationfromthecentroidofeachexistingclusterismorethan200meters.Otherwise,weaddthislocationto thecorrespondingclusterandupdateitscentroid.Finally,weconsiderallcentroidsassigni￿cantplaces. IdentifyingPlaceTypeandCharacteristics.Placescanbedescribedintermsofobjectiveinformation,such aswhetheralocationisindoororoutdoor,orinaresidentialorindustrialarea,andalsointermsofinformation thatrelatestousers’a￿ectiveappraisalsofthem.Despitepeople’sexperiencesofcoursebeingsubjective,they on average agree on a variety of characteristics such as liveliness, pleasantness, and naturalness, to name a few[17,25].Atthesametime,werecruitedcodersbelongingtothesamedemographicsoftheparticipants(e.g., studentstatus,averageage,evengenderdistribution)toreducethesubjectivenessinplaceratings.Moreover,we computedthesimilarityintheratingstoensurethattheyarereliable. Onaverage,53signi￿cantplaces(perperson)wereidenti￿edinthetwo-weekstudyperiod.Foreachparticipant weselectedthetoptenplacesinwhichtheyhadspentmosttime.Werecruitedfourindependentcoderswho wereundergraduatestudentsthemselvesbutnotparticipantsinthestudy.First,theyweretrainedinpersonand providedwithadetailedhandbookonthelocationcodingprocess.A￿ertheirtraining,theywereprovidedwith alistofcoordinates(i.e.,longitudeandlatitude)oftherespectiveplaces,andtheywereaskedtocategorizeand evaluatetheseplacesforthegivencharacteristicsbylookingatthemusingGooglemaps. Inordertocategorizetheplacetype,werelyonGoogle’sPlaceTypes[3]asthepossiblecategories.Moreover, coderswereprovidedwithanoptiontoenteranadditionalplacetypeincaseitwasnotpresentinthegivenlist. Incaseswhereaplacetypewasunclear,codersusedan“unclear”categorytodenoteanambiguousplace.Finally, eachplacetypethatwasclassi￿edbyfourcoderswasmergedbyoneoftheauthors.Itisworthnotingthatwe ￿lteredoutthecategoriesthatappearedrarelyornotatall.Inordertoperformthis￿ltering,weensuredthat placesofeachcategorywerevisitedatleastoncebyaminimumof50%oftheparticipants. Inordertoidentifythecharacteristicsofsigni￿cantplaces,thesewerealsocodedfor24descriptivecharacter- isticsthatcapturedtheambienceoftheplace,andthetypesofpeoplethatwouldvisittheplace.Inthiswork,we focusedonthefourcharacteristicratingsthatdescribetheambianceoftheplace.Speci￿cally,wefocusedonthe degreetowhichtheplacewas:inactive-busy,lazy-productive,distressing-relaxing,andnatural-urban.￿esefour characteristicswereratedona7-pointLikertscale. ￿eseratingsfromthefourcoderswerethenmergedby computingtheirmean. Finally,themergedratingswerecenteredandrescaledfroma1to7scaletoa-3to3 scale.Wethentransformthesefroma7-pointscaletoa3pointsscale(-1to1)byusingtheseranges:-3to-0.5as -1,-0.5to0.5as0,and0.5to3as1.￿isturnedthecontinuousvariableintoacategoricaloneandmakesitlikely thatwehaveenoughdataforalllevels.Levels 1,0,1representthenegative,neutral,andpositivevalueofthe � locationcharacteristicrespectively.Forinstance,theseratingsforinactive-busycharacteristicswouldconvertto inactive,neutralandbusy. 3.4.2 AppCategories. Inordertoinvestigateusers’behaviorforinteractingwithspeci￿capps,wecategorized all apps using the categories de￿ned on the Google Play store. Overall, the apps belong to 11 categories, namelyreading,￿tness,business,photography,communication,game,lifestyle,music,social,tools,andtravel applications. However,appsofcertaincategoriesarenotusedbyallparticipants. ￿erefore,weconsiderthe typeofappsthatwereusedbyallparticipants. Consequently,wecameupwiththefollowingsixcategories: communication,lifestyle,music,reading,social,andtravelapplications. PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. 1:8 • Mehrotraetal. 3.5 ￿antitativeMeasuresforNotificationandPhoneUsage In this section we discuss the metrics used in this study for quantifying users’ behavior in terms of their interactionswithnoti￿cationsandapps. Weusetwometricsforquantifyinginteractionwithnoti￿cations– Noti￿cationReceptivityandNoti￿cationSeenTime,andonemetricforinteractionwithapp–AppUsageTime. ￿eseareclassicindicatorswidelyadoptedforthistypeofstudiesbytheubiquitouscomputingcommunity(see forexample[16,30]).￿ede￿nitionsofthesemetricsarereportedbelow. Noti￿cationReceptivity:theuser’swillingnesstoreceiveanoti￿cation.￿ismetricrepresentshow • willingauseristoreceiveinterruptions. Highreceptivity(i.e.,moreclicks)indicatesincreaseinthe willingnessoftheusertobeinterruptedandviceversa.Inordertoinfertheresponsetoanoti￿cation,we checkwhetherthecorrespondingapp(whichtriggeredthenoti￿cation)waslauncheda￿ertheremoval timeofthatnoti￿cation.Weareawarethatourapproachhaslimitations,becausesomenoti￿cationsthat donotrequirefurtheractionmightnotbeclickedratherjustseenanddismissedbytheuser. Noti￿cationSeenTime(Noti￿cationAttentiveness):thetimefromthenoti￿cationarrivaluntilthe • time the noti￿cation was seen by the user. ￿is metric re￿ects the user’s a￿entiveness towards new noti￿cations.Inordertodetectthemomentatwhichanoti￿cationisseen,weusetheunlockeventof thephoneandassumethatallnewlyavailablenoti￿cationsinthenoti￿cationbarareseenwhentheuser unlocksthephone.Incaseanoti￿cationarriveswhentheuserisalreadyusingthephone(i.e.,thephone isunlocked),theseentimeofthisnoti￿cationwouldbeconsideredequaltozero.Todetectthelockand unlockeventsweusethePhoneInteractiondata(discussedinTable1).Itisworthnotingthatweremoved allnoti￿cationinstancesthatwerenotrespondedtowithin2hours.Asarecentstudy[29]demonstrated thatpeoplereceivenoti￿cationseveryhour(frommorningtolatenight),whicharehandledwithinafew minutes.￿erefore,weuse2-hourthresholdforthemaximumseentimeto￿lteroutnoti￿cationsthat arrivedwhentheuserwasawayfromthephoneorsleeping. AppUsageTime:durationforwhichanapplicationwasinforeground.Morespeci￿cally,itisthetime • intervalbetweenthelaunchofanapplicationandtheinstantoftimewhenitwassenttothebackground. We compute these metrics for each user when they are performing speci￿c activities and when they are atcertaintypesofplaces. Wethenaggregatethesemetricstocomputetheiraveragevalues. Finally,weuse statisticalteststocomparethedi￿erenceinusers’interactionwhenperformingdi￿erentactivitiesatdi￿erent typesofplaces. ItisworthnotingthatwhilecomputingAppUsageTimefordi￿erentactivities,wenormalizetheappusage valuebydividingitbythetimespentbytheuserengaginginthecorrespondingactivity.Similarly,tocompute theAppUsageTimefordi￿erenttypesofplaces,wenormalizetheappusagevaluebydividingitbythetime spentbytheuseratthecorrespondingplace.￿isstepisnecessaryinordertoavoidbiasesduetotherelative timesspentinagivenlocationorwhileengaginginacertainactivity.￿erefore,theuseofanon-normalized AppUsageTimemetriccouldproducebiasedresults. 3.6 Procedure Wewanttoinvestigatethein￿uenceofvariables,includingdailyactivities,typeandcharacteristicsofvisited locations,onthenoti￿cationandphoneusagebehaviorofusers,soweconsiderthemastheindependentvariables. Ontheotherhand,ourdependentvariablesshouldrepresentthenoti￿cationandphoneusagebehaviorofusers. ￿erefore,weusethethreenoti￿cationandphoneusagemetrics(Noti￿cationReceptivity,Noti￿cationSeenTime, andAppUsageTime)asdependentvariables. ￿eanalyseswereperformedforeachpairofindependentand dependentvariables.Weperformaone-wayANOVAforquantifyingthedi￿erencesinthedependentvariables (i.e.,Noti￿cationReceptivityandNoti￿cationSeenTime)thatrepresentsnoti￿cationinteractionbehavior.However, forquantifyingthevariabilityinthedependentvariable(i.e.,AppUsageTime)thatrepresentappusagebehavior, PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. UnderstandingtheRoleofPlacesandActivitiesonMobilePhoneInteractionandUsagePa￿erns • 1:9 ANOVA (p<0.05) 90 ANOVA (p<0.05) 40 Seen Time (mins)123000 ● ● Receptivity (%)678000 ● ● ● ● ● ● ● ● ● ● ● 50 ● 0 work sleep chores relaxation social eat exercise work sleep chores relaxation social eat exercise Daily Activity Daily Activity (a) (b) Fig.2. Roleofdailyactivityininfluencingtheuser’s(a)a￿entivenessand(b)receptivity.Di￿erentcolorindicatesstatistically significantdi￿erences(p<0.05).Intheseplotsthedotsrepresentthemeansandthebarsrepresentthestandarddeviations. Activity work chores social exercise Activity work chores social exercise sleep relaxation eat sleep relaxation eat 0.08 0.015 0.06 Density0.04 Density0.010 0.02 0.005 0.00 0.000 0 25 50 75 0 25 50 75 100 Seen Time (mins) Receptivity (a) (b) Fig.3. Probabilitydistributionsofusers’average(a)a￿entivenessand(b)receptivitywhileperformingdi￿erentdailyactivities. Here,probabilitiescanbecomputedasmultiplicationofdensitywitha￿entivenessandreceptivityvaluesrespectively. weperformatwo-wayANOVA,becauseappcategoryisconsideredasanotherindependentvariableapartfrom activityandlocationbasedindependentvariables.Itisworthnotingthatweremovedthelevelsofindependent variablesthatdidnothaveobservationsfromatleast50%oftheparticipants. 4 UNDERSTANDINGTHEROLEOFDAILYACTIVITY Inthissectionweprovideaquantitativeevaluationoftherelationshipbetweendailyactivityandtheusers’ behaviorintermsofnoti￿cationinteractionandappusage. ￿ekey￿ndingsofthissectionare: Peopleareleasta￿entivetonoti￿cationswhiletheyarepreparingtogotobed(orusingthephone • inbed)andduringthetimetheyexercise. Peoplearemorereceptivetonoti￿cationswhiletheyareexercisinganddoingchores(i.e.,routine • tasks). Overallappusageishighestwhilepeoplearerelaxingandlowestwhentheyareengagedinchores, • doingexerciseandgoingtosleep. Usageofspeci￿cappsisassociatedtousers’dailyactivities. • PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017. 1:10 • Mehrotraetal. Application Category ● communication ● lifestyle ● music ● reading ● social ● travel ANOVA App Category (p<0.05) %)3 ● Daily Activity (p<0.05) me ( ● ● Interaction (p<0.05) Ti e 2 ● g a s ● pp U1 ● ● ● ● ● ● ●● ● A0 ● ● ● ●●● ● ● ●●●● ●●● ● ●●●●● ● ● ●●●●● work eat sleep social chores relaxation exercise Daily Activity Fig.4. Relationbetweendailyactivityandapplicationusage.Inthisplotthedotsrepresentthemeansandthebarsrepresent thestandarddeviations.. 4.1 A￿entivenessandReceptivity Toinvestigatetherelationshipbetweendailyactivityandusers’a￿entivenessandreceptivitytonoti￿cations, weperformtwoseparateone-wayANOVAsthatquantifythedi￿erencesintheuser’s(i)a￿entivenessand(ii) receptivitytonoti￿cationswhiletheyperformdi￿erentactivities.￿eresultsofthe￿rstanalysis(i.e.,e￿ects ona￿entiveness)showthatthereisasigni￿cante￿ectofdailyactivitiesontheuser’sa￿entiveness,withF = 4.788,p <0.05.Inorderto￿ndwhichdailyactivitiesa￿ectusers’a￿entiveness,weperformaTukeypost-hoc test(byse￿ing� equalto0.05).AsshowninFigure2(a),thetestrevealsthattheseentimeislongest(i.e.,low a￿entiveness)whennoti￿cationsarrivewhiletheuserissleeping.However,thereisnosigni￿cantdi￿erence ina￿entivenessofuserswhiletheyareengagedinanyotherdailyactivity,exceptthattheyareslightlyless a￿entivewhileexercising. ￿eresultsofthesecondanalysisshowthatthereisalsoasigni￿cante￿ectofdailyactivitiesontheusers’ receptivitytonoti￿cations,withF =2.947,p <0.05.AsshowninFigure2(b),aTukeypost-hoctest(byse￿ing� equalto0.05)revealsthatusers’aremostreceptivetonoti￿cationswhentheyareperformingchoresorphysical exercisecomparedtootheractivities. Moreover,inordertoinvestigatethediversityinusers’a￿entivenessandreceptivity,inFigure3wepresentthe probabilitydistributionofusers’averagea￿entivenessandreceptivitywhiletheyareperformingdi￿erentdaily activities.￿eresultsdemonstratethatthereissomevariabilityinbotha￿entivenessandreceptivitybetween users.Forinstance,someusersareconsiderablylessa￿entivetonoti￿cationswhileexercisingcomparedtoother users.Users’receptivityvariesacrossallactivities. 4.2 AppUsageTime Inordertoinvestigatetherelationshipbetweendailyactivitiesandtheappusage,weperformatwo-wayANOVA byse￿ingappusagetimeasdependentvariable(DV),anddailyactivityandappcategoryasindependentvariable (IVs). Here,weusetwoIVsaswecanquantifyboththee￿ectofdailyactivityonoverallappusagetimeand thee￿ectofdailyactivityoftheuseofspeci￿capps.￿eresultsdemonstratethatalle￿ectswerestatistically signi￿cantatthe0.05signi￿cancelevel.￿emaine￿ectfordailyactivityyieldedF =7.45(p <0.05),indicatinga signi￿cantdi￿erenceintheuser’soverallappusagetime(allcategoriesconsideredtogether)whileperforming PACMonInteractive,Mobile,WearableandUbiquitousTechnologies,Vol.1,No.1,Article1.Publicationdate:July2017.