the horizon of sports is digital - dstillery · fantasy sports players to find which are most...
TRANSCRIPT
1
2019ResearchPapersCompetitionPresentedby:
TheHorizonofSportsisDigital:UsingFantasySports,e-SportsandElectronicGamblingtoFindNextGeneration
ofTicketBuyers
BusinessofSports PaperID-13378
Authors:PeterLenz,MatthewSabbanandPeterIbarra
1. Introduction Sportsteamsandleaguesareconstantlylookingfortheirnext-generationofticketbuyers.Asweprogressfurtherintothedigitalage,theexistingmethodsofconvertingfringefollowersintoticketpurchasersarelosingtheireffectiveness.Overthelastcoupledecades,largeTVcontractsandincreasedavailabilityformediaconsumptionhavemadeupforthispotentiallossinin-stadiumrevenue.[10]However,aslinearTVisincreasinglythreatenedbystreamingandover-the-top(OTT)serviceproviders[13],sportspropertieshavebeguninvestinglargeamountofresourcesintofindingnewaudiencesbasedinthedigitalenvironment;namelyFantasySports,e-Sports,ande-Gambling.[7][8]Thisresearchanalyzestheeffectivenessofthisoutreachandteststhehypothesisthatthesethreeemergingfieldsaregoodinvestmentsforsportsteamsseekingtoexpandtheircurrentpoolofin-stadiumticketbuyers.WetestthisbyoverlappingsamplesofdigitalbehaviorsbetweenNFLandMLBattendeeswithbehavioralmodelsofe-Sportsenthusiasts,e-gamblersandfantasysportsplayerstofindwhicharemostlikelytoconvertintopurchasersofin-stadiumexperiences.
2. Methodology Aspartofit’sday-to-daybusiness,Dstillerymaintainsacatalogofhundredsofbehavioralaudiencesegments[3].Sincesportfandomisadetectabledigitalsignal[5]weselectedasetofaudiencesfromDstillery’scataloguecomprisingofthreebehavioralsegments(FantasySports,e-Sports,ande-Gambling)andtwolocation-basedsegments(MLB&NFLAttendees)torepresentsportsfansandthebehaviorswewereinterestedintestingforphysicalconversionatstadiums.Weutilizedarandomsamplingprocesstodownsampleour350million-deviceuniverseintoatestaudience.Wethenperformedanagglomerativeclusteringtoidentifymutuallyexclusivesubpopulationsbasedonsimilarbehaviors.Wethencalculatedbehavioralindexprofilesforeachsubpopulationagainstnationalandseedpopulationbaselines.Theseprofilesservedasthebasisforouranalysisandconclusions.Eachstepofthemethodologyisdescribedindetailbelow. Inordertopromotereproducibilityweselectedouranalysismethodologypriortocollectinganydata.
2
2019ResearchPapersCompetitionPresentedby:
2.1.DataSources Dataforourexperimentoriginatedfromthreeprimarysources:real-timebidrequestsfromad-monetizedsites,non-monetizedwebtrafficfromthirdpartydataprovidersandappusagedatafromsoftwaredevelopmentkit(SDK)integrations.Thisuniquecombinationofdataiscrucialtoourbeingabletoaccuratelysampleonlineandmobilelocationbehaviors[2].Realtimebidrequests(BRQs)occurwhenadeviceappearsonanad-monetizedwebsiteoranapp.Whenthateventoccurs,acallissentfromthesite/apppublisherasanopportunitytofulfillanadvertisementslot.Thecallcontainsinformationsuchasanadvertisingdeviceidentifier(cookie,IDFA,AAID),atimestamp,anIPaddress,thepublisher’sname,theadcategoryandlocationdata.NotallfieldsareavailableinallinstancesofaBRQ.Thirdpartydesktopandmobileappdatastreamsareacquiredvialicensingagreementsfromapplicationswhereusershaveopted-intoprovidevisitationdatainreturnforthefunctionalityoftheapplication.Thesedatasetsallowustohaveabroaderunderstandingofonlinebehaviorbeyondad-monetizedsitesandapps.Forconvenience,weusetheterm“BRQ”torefertorecordscollectedboththroughtheRTBbidstreamandthroughSDKintegrations,asthetypeofinformationcollectedineachissimilar. Inordertomaintainacoherentviewofauseracrossmultiplescreens,wemaintainaprobabilisticnetworkofconnectionsbetweendigitallyconnecteddevices,alsoknownasadevicegraph.[2]Withthisgraph,wecandeterminewhichmobiledevicesareconnectedtowhichdesktopdevices.Thisprovidesamorerobustviewofauser’sonlinebehaviorastheyswitchdevicesandlocationsthroughouttheday. Dstilleryappliesasuiteofdataquality[4]andanti-fraud[6]filteringduringthedatacollectionprocess.Thesehavebeendescribedinasetofpublications[1][2][15]andU.S.Patents. 2.2.BehavioralModelBasedDeviceSelection Togeneratebehavioralmodels,asusedinourseedpopulation,weselectasetofwebsites(the‘seedset’)thatarehighlyindicativeofadistinctbehavior.Forexample,www.mlb.com,www.baseball-reference.com,andwww.fangraphs.comareURLswhosecontentfocusesonthecollectionandsharingofplayerstatisticsforMajorLeagueBaseball.Inbuildingabehavioralmodelonthishypothetical“Sabermetricians”seedset,oursystemcalculatespredictiveindicesforeachURLobservedintheprofilesofdevicesthatvisittheseedsetwebsites.Thetophalf-millionofthesepredictiveindicesformsthefeaturesofourbehavioralmodel.[1] Forthisexperiment,weselected10000devicesfromeachoftheFantasySports,e-Sports,ande-Gamblingbehavioralmodels. 2.3.LocationBasedDeviceSelection Location,sharedasalatitude/longitude,isoneofthekeysignalscollectedfrommobileappsthroughBRQsorourthirdpartylicensingagreements.Foreachstadiuminthisstudy,weidentifiedthecenterofthestadiumtorepresentthecentroidandcreatedageofencearoundit;translatingasinglepointintoaspecificallycraftedpolygon[15].Thegeofenceispurposefullydesignedtocontaintheentiretyofthestadiumgrounds.
3
2019ResearchPapersCompetitionPresentedby:
Figure2.1
ForeverymobileBRQandthirdpartydatapointcontaininggeospatialdata,wematchedthesharedlocationdatatoourstadiumgeofence.Ifthelocationdatafellintothepolygondimensions,wequalifiedthatdataforinclusioninourexperiment.Intheinterestofprivacyandfollowingtheautomatedlocationqualityandcontextualizationstep,therawlocationdatawasdeleted,leavingonlythepointofinterestdata,inthiscase“MLBStadium”or“NFLStadium”.[4] Werandomlyselected5000devicesobservedatNFLandMLBstadiums,duringgametimes,foracombined10000devicegroupofgameattendees.Selectingfrommultiplesportswasimportanttopreventthebehaviorsrelatedfromonesportfromdominatingtheresults.
4
2019ResearchPapersCompetitionPresentedby:
2.4.StochasticIndependentDeviceSamplingProcess Dstillery’ssystemisprogrammedtoevaluatewhichdevicescanbeusedinmodeling.OurobservationalperiodranfromAugust30,2018toSeptember15,2018thatresultedin40,070,299uniquedeviceIDswithqualifyingcharacteristics.Duetohardwareconstraints,theclusteringtechniqueoutlinedinsection2.5,canonlyconsider40000devicesandrequiresadownsampleofourtotalpopulation.Thenumberofdevicesweselectedperaudienceisoutlinedinfigure2.2.Wethenappliedapseudo-randomnumbergeneratortoeachdevicetogenerateanorderedlistofdevicesagainstthepseudo-randomnumberandselectedthefirstNdevicesforeachcharacteristic. Weusedthissametechniquetorandomlyselectanadditional40000devicesobservedduringtheobservationperiodtoserveasabaselineforcomparison.Werefertothissampleasthenationalbaseline.
Audience Type DeviceCount Comment
e-SportsFans ProbabilisticBehavioral 10000
e-GamblingFans ProbabilisticBehavioral 10000
FantasySports ProbabilisticBehavioral 10000
MLBStadiums Location-based 5000 Onlycollectedduringgametime
NFLStadiums Location-based 5000 Onlycollectedduringgametime Figure2.2
2.5.AgglomerativeClusteringinURLSpace Priortoourexperimentwecreateda128-dimensionalURLspace,structuredsuchthatdistancewithinthisspacewasrepresentativeofhowrelatedtwoURLsareintermsofsequenceofdevicevisitation.ThecloserURLSareintheembeddingspace,themoresimilartheURL’sareintheircontent.Conversely,thefartheraparttwoURLsare,thelessrelatedthetopics.[14]Tocreatethis,wetookthetimestamped,orderedhistoriesof430648822devicesandranthemthroughaConvolutedNeuralNetworkusingtheSkipgramtrainingalgorithm.ThisresultedinadatasetwhereeachinputURLhasa128-dimensionvectorthatrepresentsthatURLslocationwithintheembeddedspace. Toreallocateourseedpopulationintosubpopulationgroups,weusedeachdevice’sfullvisitationhistorytocalculatethedevice’slocationintheembeddedURLspacedescribedabove.WedidthisbyaveragingthevectorsofthevisitedURLswithinthe128-dimensionalspacetogeneralizethecontentthedeviceisinterestedin.OnceplacedintothesameembeddedURLspace,wecanusethesamedistancepropertiestounderstandthesimilarityofdevice’sbehaviors.[9] Fromthat,wecalculatedadistancematrixofeachdevicetoeveryotherdeviceintheURLspaceusingcosinesimilarity.WethenperformedanagglomerativeclusteringonthedistancematrixusingWard’smethodwithtenrandomlyselecteddevicesasclusterseeds.Thisresultedintenmutuallyexclusiveclustersofsimilarlybehavingdevices.
5
2019ResearchPapersCompetitionPresentedby:
Themethodologyofthisstepispatentpendingandaseparatemanuscriptexploringthistechniqueinfullerdepthisbeingpreparedforpublication. 2.6.Behavioralindexprofiles Weusethebehavioralaudiencesdescribedinsection2.2asasetofdescriptorsinordertounderstandanddescribeeachsubpopulation.Foragivensubpopulation,wecreateabehavioralindexprofilebycalculatingtheindexofthatsubpopulationagainstasetofbehavioralaudiencesegments,relativetoabaselinepopulation.Webuildtheseprofilesusingtwodifferentbaselinepopulations:1)theseedpopulationitselfand2)thenationalbaselinedescribedinsection2.4.InEnglish,thebehavioralindexofagivensubpopulationforabehavioralaudiencefillsintheblankinthissentence:“AdeviceinthissubpopulationisXtimesmorelikelytobeamemberofthisbehavioralaudience,comparedtoabaselinepopulation.”Thecalculationoftheindexisdescribedbelow. Webeginbyassigningthebehavioralaudiencemembershipsofeachdevicethroughthemodelingprocessdescribedinsection2.2.Theempiricalprobabilityofbehavioralaudiencemembershipisdeterminedbyexaminingsegmentmembershipsforeachuniqueuserseenwithinourseedpopulation.[2]Wethencalculatethepercentageofdevicesfromasubpopulationthataremembersofeachbehavioralsegment.Thatis,theprobabilityPofinclusioninabehavioralsegmentjforasubpopulationiisgivenby:
whereZj,iisthenumberofdevicesfoundinbothbehavioralsegmentjandsubpopulationi,andNiisthetotalnumberofusersinsubpopulationi.Thisprobabilitycanbeinterpretedasthepropensityofbehavioralaffinityjforusersinsubpopulationi. Similarly,wecalculateP(j),thebaselineprobabilityofinclusioninbehavioralsegmentj,fortherelevantbaseline.Wecanthencalculateanbehavioralindexforthisbehaviorandsubpopulationrelativetothebaseline:
Thisprocessisdonetwiceforeachsubpopulationandbehavior,onceforeachbaseline(nationalandseedpopulation).Inthisway,wecreatetwoobservationprofiles,referredtoasNationalandPopulationobservations,respectively.Byindexingagainsttheindependent,40000devicerandomsampledescribedinsection2.4,wediscoverdifferencesbetweenoursubpopulationandthenationalaudience.Byindexingsubpopulationsagainsttheseedpopulation,weareabletodiscovernuancesbetweenthesubpopulations.
6
2019ResearchPapersCompetitionPresentedby:
3. Results Theagglomerativeclusteringanalysisonthefortythousanddevicesproducedtenmutuallyexclusivesubpopulations.Thepercentbreakoutforeachsubpopulationisdisplayedinfigure3.1.The‘label’columnindicatesthesubpopulationnumber.The‘#ofDevices’columndisplaysthesumofdeviceswithinthesubpopulation.The‘PercentSize’columncalculatesthepercentofdevicesforthesubpopulation
SubpopulationLabel #ofDevices PercentSize
1 132 0.33%
2 8044 20.1%
3 524 1.3%
4 631 1.6%
5 8406 21.0%
6 4288 10.7%
7 1222 3.1%
8 1359 3.4%
9 8549 21.4%
10 6842 17.1%
Figure3.1
Visualizingtheclusteringbetweenthesubpopulationsisshowninfigure3.2.Thex-axisrepresentsthedevicesrelationshiptooneanotherasdeterminedbythemeasureddistancealongthey-axisasdescribedinsection2.5.At0.0,eachdeviceresidesinitsownspace.Theagglomerativeclusteringalgorithmcalculatesthespatialrelationshipbetweenthedevices/subpopulationsandgroupssimilarsetstogether. Inthefirstpartofouranalysis,wecalculatethesize,ofeachsubpopulationtoensureminimumthresholdsaremettoprovidestatisticallysignificantresultsinsubsequentanalysis.Subpopulations1,3and4donotmeetour1000devicethresholdrequiredforproducingsignificantresultsandaredisregardedforthepurposesofthisstudy.
7
2019ResearchPapersCompetitionPresentedby:
Figure3.2
Subpopulationswerefurtherbrokendowntounderstandthepercentofdevicesthatderivedfromeachofthefiveseed-setaudiences.Asfigure3.3shows,thepercentbyaudiencebreakoutvariedsignificantlyacrossthesubpopulations.Broadly,subpopulation2hadthemostdistinctbehavioralprofile,asitisthesubpopulationwiththehighestpercentofdevicesfromasingleseedseetaudience(e-Sports).Thissuggeststhebehavioralcompositionofthee-Sportsseed-setisdistinctfromtheotherfourseed-setaudiences(SportsGambling,DailyFantasySports,MLB&NFLAttendees).Therestofthesubpopulationsshowedamorediverseaudiencecomposition.
Subpop# NFLAttendees MLBAttendees DailyFantasySports SportsGambling e-Sports
2 1.4% 1.7% 3.6% 13.5% 79.8%
5 16.3% 16.1% 20.4% 42.6% 4.6%
6 5.2% 16.8% 33.2% 37.5% 7.3%
7 11.0% 8.3% 30.0% 46.9% 3.8%
8 33.9% 7.6% 14.4% 27.2% 16.9%
9 16.7% 17.6% 40.5% 24.1% 1.1%
10 15.2% 14.8% 24.3% 14.7% 31.0%
Figure3.3
Ourlastgoalistomeasurebehavioralsegmentparticipationratesforeachofthesubpopulationstomeasurethepotentialeffectivenessofdrivingnewticketpurchasers.
8
2019ResearchPapersCompetitionPresentedby:
4. Discussion Ineachofthebelowfigures,thenationalobservationswerederivedfromindicesrelativetothenationalaverage.Theindicesagainsttheseed-setpopulationareshownasPopulationObservations.Bothsetsofindicesareneededtoprovideinsightonthepropensitiesofasubpopulationandtounderstandthenichedifferencesbetweenthesubpopulations.ThebehaviorlabelswithintheobservationsarederivedfromDstillery’slistofpubliclyavailableaudiences.[3]
Subpopulation#
NationalObservation PopulationObservation
Subpopulation2 • VideoGames(9.91x),RolePlayingGames(9.58x)
• CollegeSports(1.01x)• FantasySports(1.0x)• ProfessionalSports(0.64x)
• JapaneseAnimeandManga(4.65x)
• CollegeSports(0.85x)• ProfessionalSports
(0.48x)
Figure4.1 Subpopulation2consistsofe-Sportsenthusiasts,gamers,andfollowersof"GeekCulture”.Theyactivelyfollowandplaycompetitivevideogamesfrommajorpublishers(AAAtitles),tabletopcardgames,andMassiveMultiplayerOnlinegames(MMOs).Thispopulationisdiverseintheirvideogaminginterests,notskewingtowardsanyparticulargenre.Inaddition,Subpopulation2isanavidenthusiastofScienceFiction/Fantasybooks,movies,andcomics.Lastly,Subpopulation2isinterestedin“Otaku”content,beingfansofJapaneseComicsandAnimation(Manga/Anime).[11] Subpopulation2Conclusion AsSubpopulation2under-indexesacrossmultiplesportsproperties,itisnotafitfordrivingnewticketpurchasers.
Subpopulation#
NationalObservation PopulationObservation
Subpopulation5 • Horseracing(5.23x)• FantasySports(2.45x)• CollegiateSports(3.33x)• Golf(3.2x),Bowling(2.14x)• Finance(1.5x),LuxuryTravel(1.2x),
CigarAficionado(1.94x)• VideoGames(0.80x)
• ConservativePolitics(1.49x)
• CollegiateSports(1.56x)
• Videogames(0.38x)
Figure4.2 Subpopulation5consistsofhardcoregamblerswhosehighestindexingactivityisHorseRacingandHorseBetting.Subpopulation5engageswithcollegiateandprofessionalsportscontentasalikelymeanstoresearchtheirgamblinginterests.Theyareinterestedineasygoingrecreationalactivitiessuchasgolfandbowling,implyinganolderdemographic.OtherinterestsofSubpopulation5includeinvestmentandfinancialnews,luxurytravel,andcigaraficionado.
9
2019ResearchPapersCompetitionPresentedby:
Subpopulation5alsoindexeshighlyagainstFantasySports,possiblyasanothermeanstogamble.Videogamesandgenerale-Sportscultureseverelyunder-indexesagainstthisgroup. Subpopulation5Conclusion Subpopulation5exhibitstraitsthatmakethemgoodcandidatesforfutureticketpurchasers.
Subpopulation# NationalObservation PopulationObservation
Subpopulation6 • Horseracing(4.86x),Gambling(1.5x)• BBQ(2.88x),CountryMusic(2.1x)• Sports(2.44x),SportsApparel(2.13x)• FantasySports(1.75x)• VacationandTravel(2.31x)• VideoGames(0.5x)
• VideoGames(0.40x)• e-Sports(0.32x)
Figure4.3 Subpopulation6consistsofgamblingenthusiaststhatdemonstratehighaffinitytowardsactivitiessuchasHorseRacing,HorseBettingandGambling.Subpopulation6followsportsatboththecollegiateandprofessionallevelandareFantasySportsenthusiasts,perhapsasmeanstofurthertheirgamblinginterests.Overall,subpopulation6exhibitsverysimilartraitstosubpopulation5however;thereislowerengagementwithinvestmentandfinanceactivities.Subpopulation6maintainsahighaffinitytowardsVacationandTravelbutunder-indexforvideogamesande-Sports. Subpopulation6Conclusion Subpopulation6isagoodcandidateforin-stadiumticketpurchasers.
Subpopulation# NationalObservation PopulationObservation
Subpopulation7 • Hunting/Trapping(10.0x)• Trucking(9.26x)• Guns(9.2x)• Fishing(8.67x)• CountryLife(8.33x)• Sports(2.33x)• FantasySports(0.74x)• VideoGames(0.63x)
• Trucking(7.39x)• Hunting/Trapping(7.0x)• PowerTools(6.45x)
Figure4.4 Subpopulation7isinterestedinSportsandoutdoors,over-indexingoncontentrelatedtohunting,off-roadtruckingandfishing.Theyarefansofvarioussportspropertiesatboththecollegiateandprofessionallevelbutunder-indexforFantasySports,videogamesande-Sports.Similartosubpopulations5and6,gamblingindexeshighlyforthisaudience.
10
2019ResearchPapersCompetitionPresentedby:
Subpopulation7Conclusion Asshowninfigure3.3,subpopulation7doesnotconsistofmanyMLB/NFLattendeesdespitetheinterestandfandomofsportentities.Duetothesubpopulation’ssmallsize,itisinconclusivewhethertheyareprimecandidatesforticketbuying.
Subpopulation#
NationalObservation PopulationObservation
Subpopulation8 • Sneakers(8.98x),CelebrityNews(6.12x),Boxing(5.78x)
• Sports(5.53x)• FantasySports(3.34x)• VideoGames(1.25x)• E-Sports(0.99x)
• Sports(7.95x)• Sneakers(6.85x)• CelebrityNews
(6.3x)
Figure4.5 Subpopulation8showsinterestinSportsandFantasySportsmorethantheaverageconsumer.TheyareinterestedinVideoGamesbutnotthetypesofgamesthatareprevalentamonge-Sports,asseeninsubpopulation2.Thedistinguishingfeatureofthisgroupistheover-indexingforsneakerandcelebritynews,suggestingmoremainstreamconsumertendencies. Subpopulation8Conclusion Whilesubpopulation8consistsofattendeestoNFLeventsandisfansoflocalsportsteam,itisinconclusiveonwhetherornotthispopulationisidealforbecomingticketpurchasersduetoitssize.
Subpopulation# NationalObservation PopulationObservation
Subpopulation9 • FantasySports(5.96x)• ProfessionalSports(3.71)• CollegiateSports(4.33x)• Finance(2.53x),Investment(2.32x)• LuxuryTravel(1.43x)• VideoGames(1.0x)
• Finance(3.11x)• Sports(2.32x)
Figure4.6 Subpopulation9consistsofdiehardFantasySportsfansindexingatthehighestamongallthesubpopulations.Thisgroupfollowssports,fromthelocaltonationallevel,atthehighestpropensity.Similartosubpopulation5,thisgroupshowsahighaffinitytowardsfinanceandinvestment,indicatingdisposableincomeforvariousactivities.Uniquetothissubpopulation,thereisoverlappingpropensityforbothsportsandcasualvideogamebehaviors. Subpopulation9ConclusionThehigh-indexinginterestinsportspropertiesandfantasysportsmakethissubpopulationaprimecandidateforinvestment.
11
2019ResearchPapersCompetitionPresentedby:
Subpopulation#
NationalObservation PopulationObservation
Subpopulation10
• VideoGames(2.24x)• Libertarian(3.22x),
Conservative(2.99x)• FantasySports(2.18x)• Sports(2.08x)• MMOs(1.79x)
• Computers,DIY/ComputerParts(1.5x)
• Sports(1.08x),Fantasysports(0.85x)
Figure4.7 Subpopulation10playsvideogamesmorethantheaverageuser.TheysticktoMassivelyMultiplayerOnlinegames(MMOs)orotherbigtitlevideogamesfrommajorpublishersbutarenotnecessarilythehardcorecompetitivegamersobservedinsubpopulation2.ThisaudienceisfullofenthusiastsintheInformationTechnologysector,researchingthelatesthardwarefortheirgamingsystems.Affinitiesindicatethatsubpopulation10’sprefersPCgamingbutisalsoconsumersofhomevideogameconsoles.Comparedtothenationalaverage,subpopulation10istechnicallysavvyandavidfollowersofpolitics. Subpopulation10Conclusion Inrelationtosportsproperties,theseusershaveinterestinFantasySportsandgeneralSportsnews,at1.5xthenationalindex.Resultsshowsubpopulation10indexeshighlyfortheirlocalsportsteamsandisanidealaudienceforbecomingticketpurchasers.
5. Conclusion Insummary,thee-Sportscentricsubpopulation2showedthelowestbehavioraloverlapwithin-stadiumattendees.Basedonthisresearch,wedonotbelievee-Sportsisanidealchannelforfindingfutureticketpurchasers.Incontrast,subpopulations5-10resultedinadiversebehavioralcompositionoffantasysports,e-gamblers,andin-stadiumattendees.Weconcludethate-gambling,asdemonstratedinsubpopulations5and6,isthebestaudienceforinvestmentinfindingfuturein-stadiumticketpurchaserswithFantasySportsbeingaviablealternative. Wepresentaforward-lookingmethodologyintohowteamsshouldspendtheirinvestmentresourcesandexpandtheunderstandingofhowanewmarketcanbetargetedwithpreciseandeffectivemessaging.Anin-depthunderstandingofapotentialacquisitionaudience,beforetheallocationofmoneyandresources,willonlyincreasethelikelihoodofsuccess.Webelieveourexperimentisafirststepinsheddinglightontoanareaofinvestmentthatisdifficulttoquantify.Finally,weenvisionapplicationsbeyondthestudyofaudiences.We’vedemonstratedthemodel’sabilitytomeasuretheinterestsofasubpopulationwithouttheneedforfirstpartydatasources.Addingadditionaldatasources,suchaspurchasedata,wouldonlyincreasetheeffectivenessofthismethodologyandallowustoprovideincreasedspecificityintheanalysisofresults.
12
2019ResearchPapersCompetitionPresentedby:
6. Acknowledgement WewouldliketothankDstilleryand,mostespecially,theentireDstilleryDataScienceteam,includingourChiefDataScientist,Dr.MelindaHanWilliams,fortheirinvaluablefeedbackandinsight.Theirwillingnesstoassistandansweranyquestionswehadinthisprojectensuredourresultswereofthehighestquality.WealsothankAbbyBeltraniforherinsightintothebusinessofsportsmarketingindustry.Lastly,wethankourfamiliesfortheirconstantandunwaveringsupport.
References 1. Wojan,Timothy,Crown,Daniel,Slaper,Timothy,Lenz,Peter,Bianco,Alyssa.“Arethe
ProblemSpacesofEconomicActorsIncreasinglyVirtual?WhatGeo-locatedWebActivityMightTellUsaboutEconomicDynamism”.65thAnnualMeetingsoftheNorthAmericanRegionalScienceCouncil.NARSC,2018.
2. Slaper,Timothy,AlyssaBianco,andPeterLenz."DigitalVaporTrails:UsingWebsiteBehaviortoNowcastEntrepreneurialActivity."2ndInternationalConferenceonAdvancedResearchMethodsandAnalytics(CARMA2018).EditorialUniversitatPolitècnicadeValència,2018.
3. “FULLAUDIENCETAXONOMY“dstillery.comDstillery,2018.https://audiences.dstillery.com/taxonomy
4. Williams,MelindaHan,OriM.Stitelman,andRodneyAlanHook."Evaluatingauthenticityofgeographicdataassociatedwithmediarequests."U.S.PatentNo.9,948,733.17Apr.2018.
5. Kim,JoonK.,andKevinHull."HowfansareengagingwithbaseballteamsdemonstratingmultipleobjectivesonInstagram."Sport,BusinessandManagement:AnInternationalJournal7.22017:216-232.
6. Stitelman,OriM.,etal."Methods,systemsandmediafordetectingnon-intendedtrafficusingco-visitationinformation."U.S.PatentNo.8,719,934.6May2014.
7. Fischer,Eric“MillennialsPutTicketStrategiesToTest”SportsBusinessDaily,2015.https://www.sportsbusinessdaily.com/Journal/Issues/2015/06/08/In-Depth/Ticketing-main.aspx
8. Shaikin,Bill,Mitchell,Houston.“MLBBecomesThirdMajorSportsLeagueToFormPartnershipwithMGM”LosAngelesTimes,2018.https://www.latimes.com/sports/mlb/la-sp-mlb-mgm-partnership-20181127-story.html
9. Arora,Sanjeev,YingyuLiang,andTengyuMa."Asimplebuttough-to-beatbaselineforsentenceembeddings."InternationalConferenceonLearningRepresentations,ICLR,2017.
10. Leitch,Will.“Nobody'sGoingToSportsinPersonAnymore.AndNoOneSeemsToCare.”NewYorkerMagazine,2018.https://nymag.com/intelligencer/2018/07/nobodys-going-to-sports-in-person-and-no-one-seems-to-care.html
11. Hiroki,Azuma,JonathanE.Abel,andShionKono."Otaku:Japan’sdatabaseanimals."Trans.JonathanE.AbelandShionKono.Minneapolis:UniversityofMinnesotaPress,2009.
12. Dalessandro,Brian,etal."Scalablehands-freetransferlearningforonlineadvertising."Proceedingsofthe20thACMSIGKDDinternationalconferenceonKnowledgediscoveryanddatamining.ACM,2014.
13. Keating,Cal."OvertheToporovertheHeadsofSportsBroadcasting:SportsandEntertainmentContentLicensingandDistributioninaNewEra."SportsLaw.J.25,2018:177.
13
2019ResearchPapersCompetitionPresentedby:
14. Abadi,Martín,etal."Tensorflow:asystemforlarge-scalemachinelearning."Proceedingsofthe12thUSENIXSymposiumonOperatingSystemsDesignandImplementation(OSDI).Vol.16.2016.
15. Lenz,Peter,Ibarra,Peter."UsingDigitalSignalsToMeasureAudienceBrandEngagementAtMajorSportsEvents:The2015MLBSeason."10thMIT-SloanSportsAnalyticsConference.MIT,2016.