Social media and crisis management Chiara Francalanci WSIS, Geneva May 5, 2016
Socialmediaandcrisismanagement
ChiaraFrancalanci
WSIS,GenevaMay5,2016
Socialmedia
• RepresentfastandeffecBvecommunicaBonchannels.
• Allowtheunsolicitedexpressionofpersonalviews.
• ProvideanopendisplayofconnecBons.• EnablesocialmediaanalyBcs,withavarietyofapplicaBons.
• Enablecrowdsourcing.
Asocialmediaparadigm:Crowdsourcing
• ReferencemodeltoidenBfyandmanagesharedissues
• SoluBonsdelegatedtothecrowd• Trustonthe«wisdomofthecrowd»
SocialnetworksrepresentthelargestglobalcommuniBes.Canweleveragetheirstrengths?
TORCIAProject–ObjecBves
TodesignaplaUormthatsupportsthereal-BmeaccesstoTwiWerinformaBon,byselecBngdependableposts, by spreading importantmessages, andby allowingthecooperaBonbetweeninsBtuBonsandthecrowd.InnovaBvetechnologymodules:1) asemanBcenginethathelpsinformaBonmanagementandprovidesalarms
andtriggers.2) amobile app that represents the virtual cooperaBon environment (under
development)
RecentusecasesFloodsinSardegna–Sept.2013:• Over30Ktweetsontheflood• UseofsocialnetworkstogetrealBme
informaBon(#allertameteoSAR)• Useofsocialnetworkstocoordinate
recoveryacBviBes(whohasroomfordisplacedpeople)
Emergenza24:• UseofSocialNetworksfor
EmergencyManagement• LimitedtoTwiWerwithprecise
guidelines(#Emergenza24)
13911
73696079 6237
105709714 9278
8130
11999
7245
37392449 2360
57114973 4724
3725
6530
0
2000
4000
6000
8000
10000
12000
14000
16000
# T
wee
t
10 day intervals
# Tweet totali
# Tweet classificati
VolumesinItalianfromDec.2012toFeb.2013• AveragevolumesinItalian:40.000tweet/month• OnlyhalfofthepostsarerelatedtofloodsaderdisambiguaBon• VolumesinEnglisharetenBmeshigher
SocialmedianalyBcs–volumesofbuzz
Postsusefulduringtheresponsephasearepredominant
0
1000
2000
3000
4000
5000
6000
7000
8000
Num
ber
of t
wee
ts c
lass
ified
10 day intervals
Preparedness
Response
Recovery
Mitigation
SocialmediaanalyBcs–topicsandphases
TesBng–Sardiniaflood11/18/2013
OtherExamples:Predictingsaleswithsentiment(telefonino.net)
Weeksbetween1/1/2012and30/4/2012.Overlappingyaxes.
Other examples: predic1ng financial risk with sensi1ve news
Service:dailyfeedofnewswithkeyinformaBonextractedaccordingtoapredefinedformatSources:ItalianlocalnewspapersandlocalnewssitesSubjects:newsaboutcompaniesandanykindofcommercialacBvity.Thecompany’snameisextractedsemanBcallyfromthenews.Thatis,thecompany’snameisNOTextractedbymatchingthenewsagainstadatabaseofcompanynames.Typeofnews:weprovidenewswherethecompanyisassociatedwithcriBcaleventssuchasfurto,incendio,multa,cassaintegrazione…(200+typesofcriBcaleventavailable)RaBoofsignificantnews:foreachsignificantnewsprovidedinthefeed,about1000newsareanalyzed(raBo=1/1000)
Predic1ng financial risk Sensi1ve news: Example
SESTO-UnincendiosièsviluppatonellanoWeinuncapannoneinviaFerminellazonadell'Osmannoro,aSestoFiorenBno.LefiammehannointeressatometàdellastruWura,circa1.000metriquadraB,cheospitaunadiWadipelleWeriagesBtadaunciWadinocinese.IlrogohacausatoilcrollodelteWo.Ivigilidelfuoco,intervenuBcondieciautomezzietrentauominidaFirenzeePrato,sonoancorasulposto:lefiammealmomentosonosoWocontrollo.Secondoquantoemerso,ilrogo,divampatoquandononc'eranessunoall'internodelladiWa,sarebberoscaturitepercauseaccidentali,forseuncortocircuito.Inbaseadunaprimavalutazionedeivigilidelfuoco,cheancoranonsonoentraBnellastruWuraacausadellealtetemperature,circametàdelcapannone,dovehasedeunacasaeditrice,nonsarebbestatainteressatadallefiamme,chesarebberostatecontenutedaunmurotagliafuoco.27novembre2012.
streetaddresscriBcaleventcity
typeofcompany
date
• ThesemanBcengineextractstheelementsavailableinthenews
• Intheexample,thecompanynameisnotreportedinthenews
• Thecompanynameisobtainedinthesubsequentmatchingphase,byqueryingdatabasesofItaliancompanies,providingtheelementsextractedfromthenewsassearchkeys
• Theresultofthematchingphaseis:• Companyname:ChengXiangS.r.L.• Address:ViaEnricoFermi50/52,50019
SestoFiorenBno(FI)• Otherdata,suchastheParBtaIva
numberarealsoprovided
Socialmediaandsecurity
• OpportuniBesaremaximizedifusersareregisteredandtheironlineidenBtyisknownacrossdifferentsocialmedia
• Inothercontexts,registraBonisnotpossible,butwhensecurityisthegoalpre-registraBonseemsaviablesoluBon
• Pre-registraBonwouldenabletheanalysisofsocialmediainformaBontotraceuserbehaviorandidenBfybehavioralpaWernsthatarecorrelatedwithspecificsecuritythreats.
• PreviousexperienceinthesefieldsshowsthatsocialmediaareavalidcomplementtomoretradiBonalsourcesofinformaBon.
AneedforasemanBcmodelThebasicbuildingblocksofasemanBcmodelforsocialmediaanalyBcsare:• A“securityissue”whichistheprimarytopicofthemodel,and
“categories”whicharesub-topicsunderthatissue.Forexample,“recruitment”couldbeacategoryundertheissueof“terrorism”.
• DefiniBonsofenBBesthatarespecifictothesecurityissue,includingpropernames(e.g.espionage,organizedcrime,criminalgang,orISIS).
• Mappingofthelanguage,expressionsandtypicalsyntacBcstructureusedtoindicatespecificacBonsoropinions(inmanycase,alternaBvemeaningsareassociatedwithmainstreamconversaBonpaWern,afocusonthespeakerhelps).
• BuildataxonomyforsenBmentanalysis(e.g.“makethempay”).
Conclusions
• SocialmediaanalyBcsraiseabigdataissue• MulBplesocialmediashouldbeconsideredandtheidenBtyofpeopleshouldbetracedacrossmedia
• DomainexperBseiskeytobuildasemanBcmodel
• PredicBveanalyBcsrequirehistoricaldataoverextendedperiodsofBme