Sen imen and Linguis ic Analysis o Epidemic
Ou b eak Da a om O icial and Al e na i e
Sou ces
Ka ina O doñez Gue e o
Facul ad de Posg ado
Uni e sidad Técnica Es a al de
Que edo
Que edo, Ecuado
[email p o ec ed]
ORCID: 0009-0009-2507-0519
José Co de o Bazu o
Facul ad de Posg ado
Uni e sidad Técnica Es a al de
Que edo
Que edo, Ecuado
[email p o ec ed]
ORCID: 0009-0001-2961-6736
Edua do Samaniego Mena
Facul ad de Posg ado
Uni e sidad Técnica Es a al de
Que edo
Que edo, Ecuado
esamanie[email p o ec ed]
ORCID: 0000-0002-6196-2014
Geo anny B i o Casano a
Facul ad de Posg ado
Uni e sidad Técnica Es a al de
Que edo
Que edo, Ecuado
[email p o ec ed]
ORCID: 0000-0002-7715-7706
Abs ac — In o ma ion on epidemic ou b eaks is a key inpu o
heal h su eillance, as i allows o he assessmen o he sp ead and
associa ed social pe cep ion. This s udy examines emo ional and
linguis ic pa e ns in na a i es dissemina ed by in e na ional
o ganiza ions (WHO, UN, CDC) and digi al pla o ms (Google
News and Reddi ) o e a h ee-mon h pe iod. The KDD p ocess was
applied in R S udio (selec ion, p ep ocessing, ans o ma ion,
modeling, and e alua ion), using Bing and NRC lexicons and a
supe ised Nai e Bayes model o enhance he de ec ion o
emo ional nuances. A o al o 12,340 ex s (3,100 om o icial
sou ces, 4,240 om Google News, and 5,000 om Reddi ) we e
analyzed using s anda dized que ies in English: pandemic,
con inemen , epidemic, and HMPV. O icial sou ces showed a
g ea e p esence o posi i e emo ions linked o coope a ion and
secu i y; Google News concen a ed nega i e na a i es wi h e ms
such as isk and dange ous; Reddi combined ea and sadness wi h
appea ances o hope. The analysis included - es s and ANOVA
wi h 95% con idence in e als. The wo k is explo a o y and
p elimina y in na u e and sugges s ha su eillance sys ems should
in eg a e he moni o ing o social ne wo ks and digi al media, along
wi h public policy measu es o imp o e communica ion in heal h
c isis si ua ions.
Keywo ds— epidemic ou b eaks, sen imen analysis, ex
mining, epidemiological su eillance, public communica ion.
I. INTRODUCTION
The analysis o in o ma ion on epidemic ou b eaks is a
key esou ce o public heal h, especially in a global scena io
whe e diseases ha e he capaci y o sp ead apidly be ween
coun ies and con inen s [1]. In addi ion o he
epidemiological cou se, a p ac ical p oblem is how na a i es
in luence public con idence and he adop ion o measu es
(e.g., adhe ence o ecommenda ions o accep ance o
accines). The he e ogenei y o he da a poses challenges
ela ed o i s eliabili y, consis ency, and ele ance in
esponding o immedia e needs [2]. Iden i ying bo h he
s eng hs and limi a ions o hese sou ces helps o e ine
su eillance sys ems and guide communica ion s a egies
du ing heal h eme gencies [3].
O icial sou ces, ep esen ed by he Wo ld Heal h
O ganiza ion (WHO), he Uni ed Na ions (UN), and he
Cen e s o Disease Con ol and P e en ion (CDC), we e
selec ed due o hei ecogni ion in p o iding alida ed and
s uc u ed in o ma ion used by These ins i u ions apply
igo ous p o ocols in he collec ion and dissemina ion o da a;
howe e , hei immediacy may be limi ed by adminis a i e
p ocesses ha delay eal- ime upda es [5]. In con as , digi al
pla o ms such as Google News and Reddi unc ion as
al e na i e sou ces ha allow access o mo e apidly
dissemina ed na a i es and spon aneously exp essed
collec i e pe cep ions. The use o hese sou ces ca ies isks,
such as he ci cula ion o incomple e o biased in o ma ion
[6]. [7]
The choice o Google News and Reddi was based on
c i e ia o co e age and na a i e di e si y. Google News, as
a global news agg ega o , compiles publica ions om a ious
in e na ional media ou le s, while Reddi ocuses on
discussions in hema ic communi ies such as /epidemiology
and /heal h. S anda dized que ies wi h English keywo ds
(pandemic, con inemen , epidemic, HMPV) we e applied o
bo h pla o ms, ensu ing consis ency and acili a ing
compa isons be ween sou ces. Howe e , i is ecognized ha
he inclusion o hese pla o ms in oduces biases inhe en o
he na u e o he con en : in Google News, due o he media's
ocus on cap u ing a en ion, and in Reddi , due o he
in luence o communi ies wi h pa icula in e es s.
The esea ch uses he KDD (Knowledge Disco e y in
Da abases) p ocess, s uc u ed in phases o selec ion,
p ep ocessing, ans o ma ion, modeling, and e alua ion. In
addi ion o sen imen lexicons (Bing and NRC), a supe ised
ex classi ica ion model was inco po a ed o expand he
de ec ion o emo ional nuances and e alua e di e ences
be ween sou ces. Sen imen analysis p o ides use ul signals
o public heal h by allowing he de ec ion o social ala m
peaks, moni o ing changes in one, and p io i izing isk
messages.
The main pu pose o his s udy is o compa e emo ional
and linguis ic pa e ns in na a i es o epidemic ou b eaks
om o icial and al e na i e sou ces, analyzing a iables such
as speed o dissemina ion, geog aphical co e age, and
emo ional cha ge. The indings desc ibe how isks and
expec a ions a e communica ed in di e en se ings and o e
applicable inpu s o s eng hening su eillance sys ems and
designing public policy ecommenda ions aimed a imp o ing
global heal h communica ion. [8] [9], [10]
II. RELATED WORK
The collec ion and analysis o in o ma ion on epidemic
ou b eaks om a ious sou ces has enabled ad ances in he
de ec ion, moni o ing, and managemen o diseases globally.
Access o o icial sou ces, news, and social media pla o ms
has ans o med he way da a on disease sp ead is collec ed,
enabling a g ea e di e si y o analy ical app oaches [11].
Below, we de ail how hese sou ces ha e been used in
p e ious esea ch, highligh ing he applica ion o da a science
o op imize esul s and o e come challenges inhe en in
managing la ge olumes o in o ma ion.
O icial sou ces, such as he WHO, he UN, and he CDC,
ha e p o en o be undamen al ools in consolida ing
epidemiological da a based on o mal epo s. Fo example,
[12] desc ibes he use o isual ools o explo e pa e ns o
epidemic sp ead in spa ial and empo al dimensions, using
da a om eliable sou ces o assess he dynamics o diseases
such as COVID-19. This app oach has p o en unc ional in
modeling complex scena ios ha equi e p ecision in he
ep esen a ion o ou b eaks.
A he same ime, epidemic p opaga ion models ha e been
analyzed using wo-laye ne wo ks, whe e physical and i ual
connec ions allow he sp ead o in o ma ion and diseases o
be simula ed [13]. This app oach demons a es how o icial
da a can be in eg a ed wi h complex simula ions o assess he
impac o connec i i y on he sp ead o diseases.
In [14], i is documen ed how ex analysis using ad anced
na u al language p ocessing echniques has made i possible
o iden i y ends in he ea ly s ages o epidemic ou b eaks,
he eby imp o ing he abili y o o icial sys ems o adap o
changing scena ios.
O icial sou ces and al e na i e sou ces, such as social
media, play a undamen al ole in he collec ion and analysis
o da a on epidemic ou b eaks. In [15], he au ho s e alua e
how he sp ead o p e en i e in o ma ion, whe he posi i e o
nega i e, on mul iplex ne wo ks can in luence public
pe cep ion and esponses o ou b eaks. These indings
highligh he impo ance o conside ing in o ma ion dynamics
in he o mula ion o heal h su eillance and esponse
s a egies.
The compa a i e analysis be ween o icial and al e na i e
sou ces has also been add essed in [16], whe e he
p obabili ies o ex eme epidemics occu ing we e analyzed
using his o ical and cu en da abases. This s udy highligh s
how he in eg a ion o mul iple sou ces imp o es he abili y o
p edic and espond o la ge-scale e en s.
In ecen yea s, social media has eme ged as an al e na i e
sou ce o de ec ing epidemic ou b eaks. Fo example, [17]
analyzes how he sp ead o in o ma ion on digi al pla o ms
can se e as an ea ly indica o o heal h e en s. Thei esea ch
highligh s he use o op imiza ion algo i hms o iden i y
c i ical nodes in hese ne wo ks, showing how al e na i e
sou ces complemen he limi a ions o o icial sou ces by
cap u ing collec i e pe cep ions and beha io pa e ns in eal
ime. Ano he s udy, desc ibed in [18], in es iga ed he
heo e ical limi s o in e ence in epidemic sp ead h ough
s a is ical mechanics me hods applied o social ne wo ks and
g aphical models. Thei wo k highligh s how use -gene a ed
da a can p o ide addi ional insigh s when in eg a ed wi h
adi ional models.
Reddi , in pa icula , has p o en o be an e ec i e sou ce
o de ec ing eme ging ends in public pe cep ion. The
analysis o pos s on sub eddi s such as /epidemiology has
been documen ed in [19], indica ing ha hey used dis ibu ed
con ol models o assess how decen alized ne wo ks can
in luence he sp ead o in o ma ion and diseases.
The in eg a ion o da a science has been undamen al in
add essing he limi a ions inhe en in he he e ogenei y o
in o ma ion sou ces. Fo example, in [20], hey de eloped a
model o classi y and analyze epidemic in o ma ion ex ac ed
om social ne wo ks, achie ing high le els o classi ica ion
accu acy and gene a ing concise summa ies o pos s.
Addi ionally, he wo k in [21] employed ad anced
me hods o in e epidemic ajec o ies om pa ial
obse a ions, using Bayesian echniques o imp o e he
es ima ion o dynamic pa ame e s. Such s udies show how
da a science-based models can add ess challenges ela ed o
unce ain y and a iabili y in da a.
III. METHODOLOGY
The me hodology p oposed in his s udy was designed o
analyze in o ma ion on epidemic ou b eaks om o icial and
al e na i e sou ces, inco po a ing ex mining, s a is ical
analysis, and isualiza ion echniques. A sys ema ic p ocess
was adop ed ha included he s ages o iden i ica ion,
collec ion, cleaning, analysis, and compa ison o da a, wi h
he aim o ensu ing he accu acy o he esul s ob ained.
A. S udy Design
The s udy is based on a compa a i e app oach ha
combines au oma ed da a consump ion echniques and
ad anced analysis ools o examine cha ac e is ics and
di e ences be ween in o ma ion sou ces. These include
o icial pla o ms, such as he WHO, UN, and CDC websi es,
along wi h al e na i e sou ces such as Google News and social
media, speci ically Reddi . The keywo ds pandemic,
con inemen , epidemic, and HMPV we e applied uni o mly
ac oss all sou ces, wi h he aim o explo ing aspec s such as
he speed o da a dissemina ion, geog aphic co e age, and he
p edominan emo ions in he epo ed na a i es.
B. Popula ion and Sample
The esea ch wo ked wi h a o al o 12,340 ex s,
dis ibu ed ac oss 3,100 eco ds om he WHO, UN, and
CDC, 4,240 a icles om Google News, and 5,000 pos s on
Reddi . This sample was ob ained o e a h ee-mon h pe iod
wi h pe iodic collec ions ha acili a ed he obse a ion o
empo al a ia ions.
1) O icial Sou ces:
These include globally ecognized o ganiza ions such as
he Wo ld Heal h O ganiza ion (WHO), he Uni ed Na ions
(UN), and he Cen e s o Disease Con ol and P e en ion
(CDC). These en i ies we e selec ed due o hei abili y o
gene a e eliable epo s, eal- ime upda es, and de ailed
analyses o disease sp ead. The in o ma ion p o ided by hese
sou ces includes s a is ics, heal h ale s, and global epo s.
2) Al e na i e Sou ces:
This ca ego y included Google News and he social
ne wo k Reddi . Google News was used o access in o ma ion
o in e es published by in e na ional news media h ough
speci ic que ies ela ed o epidemic ou b eaks. Reddi , o i s
pa , o e s an o ganized s uc u e in hema ic sub eddi s such
as /epidemiology, /heal h, and /wo ldnews, which allow o
he iden i ica ion o publica ions wi h echnical con en and
social pe cep ions abou heal h e en s.
C. Da a collec ion ool
The ool designed o his esea ch was in ended o cap u e
s uc u ed in o ma ion abou epidemic ou b eaks. Uni o m
que ies using he keywo ds pandemic, con inemen , epidemic,
and HMPV we e used in all selec ed sou ces (o icial and
al e na i e). The analysis was pe o med in R S udio wi h
lib a ies such as es , h , jsonli e, idy ex , and ggplo 2,
which acili a ed ex ac ion, p ocessing, and isualiza ion.
Au oma ed que ies we e con igu ed o collec in o ma ion
di ec ly om a ailable APIs and po als, which allowed o
s anda diza ion o sea ch c i e ia and ensu ed aceabili y. On
Reddi , access was p o ided h ough he Reddi API wi h
OAu h2 au hen ica ion and JSON eques s managed wi h
h /jsonli e o sea ch endpoin s and each sub eddi ; quo as
and usage policies we e espec ed and no sc aping was used.
Fo Google News, compa ible public eeds (RSS/JSON
o ma ) accessible om o icial agg ega o links we e used.
D. Collec ion p ocess
The collec ion was ca ied ou o e a pe iod o h ee
mon hs a egula in e als, allowing o he cap u e o
empo al a ia ions. Fo Google News, que ies we e de ined
using English-language keywo ds, il e ing esul s in o heal h
and science ca ego ies. On Reddi , he same keywo ds we e
applied o speci ic sub eddi s ( /epidemiology, /heal h,
/wo ldnews), wi h il e s by language and emo al o URLs,
emojis, and duplica e con en .
I is ecognized ha hese sou ces in oduce biases: in he
case o Google News, due o he media's endency owa d
impac ul na a i es, and on Reddi , due o he in luence o
communi ies wi h pa icula in e es s. Technical limi a ions:
changes in API endpoin s o quo as, in e mi en a ailabili y
o eeds, co e age mainly in English ha may exclude local
nuances, and lack o access o con en ha was dele ed,
p i a e, o mode a ed be o e cap u e.
The collec ed da a was o ganized in o homogeneous
s uc u es, and andom checks we e applied o e i y
consis ency wi h he o iginal sou ces. This alida ion ensu ed
ha he eco ds we e comple e and consis en wi h he s udy
objec i es.
E. Da a Analysis
Da a analysis was ca ied ou ollowing he i e phases o
he Knowledge Disco e y in Da abases (KDD) p ocess,
summa ized in he diag am in Figu e 1 (Sou ce selec ion →
Da a collec ion → P ep ocessing → T ans o ma ion →
Modeling & analysis → E alua ion & isualiza ion). This
p ocedu e allowed o he sys ema ic o ganiza ion o he
ex ac ion, cleaning, ans o ma ion, modeling, and e alua ion
o in o ma ion om o icial and al e na i e sou ces. The
implemen a ion was ca ied ou in R S udio using lib a ies
such as es , h , jsonli e, dply , idy ex , ggplo 2,
wo dcloud2, s ing , g idEx a, idy , and ex da a, which
acili a ed he manipula ion, p ocessing, and isualiza ion o
ex ual eco ds.
Fig. 1. KDD pipeline used in he s udy
1) Da a Selec ion:
Reliable sou ces we e selec ed o he esea ch. O icial
pla o ms included he WHO, UN, and CDC, ecognized o
p o iding up- o-da e epidemiological da a. Google News and
Reddi we e conside ed as al e na i e sou ces. In Google
News, news ela ed o speci ic keywo ds was analyzed, while
in Reddi , hema ic sub eddi s such as /epidemiology,
/heal h, and /wo ldnews we e explo ed. These pla o ms
p o ided in o ma ion on public pe cep ions and na a i es
ela ed o epidemic ou b eaks.
2) Da a P ep ocessing:
Cleaning included emo ing duplica es, no malizing
da es, and co ec ing ypos. Tex s we e con e ed o lowe case
and il e s we e applied o exclude incomple e en ies. The
dply and s ing lib a ies we e used, ensu ing ha he inal
da abase was aligned wi h he objec i es o he analysis.
3) Da a T ans o ma ion:
The cleaned eco ds we e o ganized in o s uc u es
compa ible wi h ex ual and isual analysis. Tidy was used o
s uc u e he co pus and ex da a o associa e emo ions wi h
he na a i es. In addi ion o he Bing and NRC lexicons, a
supe ised emo ional classi ica ion model wi h c oss-
alida ion was ained, which allowed o he de ec ion o
nuances no co e ed in basic dic iona ies. This in eg a ion
acili a ed he iden i ica ion o mixed emo ions in he
discou ses.
4) Modeling and Analysis:
Na a i e pa e ns we e ep esen ed using ggplo 2 and
wo dcloud2, gene a ing wo d clouds, ba cha s, and sca e
plo s. All isualiza ions we e p oduced in English ( i les, axes,
and legends), wi h a minimum esolu ion o 300 dpi and on
size ≥ 10 p o ensu e legibili y. Figu e 2 includes a colo ba
wi h an explici scale: –1 = nega i e emo ion, 0 = neu al, +1
= posi i e emo ion; he ends o he g adien co espond o
hose limi s. The analysis included he calcula ion o a e age
sen imen sco es pe sou ce, as well as he applica ion o
S uden 's - es s and ANOVA o con as di e ences be ween
g oups, wi h 95% con idence in e als.
5) Resul s E alua ion
The e alua ion o he collec ed and p ocessed da a was
ca ied ou h ough a compa a i e analysis o o icial and
al e na i e sou ces, ocusing on aspec s such as he accu acy,
consis ency, and ep esen a i eness o he in o ma ion. This
p ocess made i possible o e i y he quali y o he da a and
ensu e i s alignmen wi h he esea ch objec i es. To his end,
se e al me hodological aspec s we e e alua ed:
• Accu acy o in o ma ion: C oss-checks we e
pe o med wi h he o iginal epo s om each
pla o m, con i ming ha he da a ex ac ed h ough
au oma ed que ies co esponded o e i iable
na a i es. This alida ion was use ul o disca ding
duplica e eco ds o hose ou side he de ined ime
ange.
• Consis ency be ween sou ces: Emo ional and
hema ic pa e ns we e analyzed o de e mine he
co espondence o esul s wi h he ini ial que ies
(pandemic, con inemen , epidemic, HMPV). The
con as be ween na a i es showed di e ences linked
o he biases o each sou ce: a endency owa d high-
impac con en on Google News and a p edominance
o communi y pe spec i es on Reddi .
• Da a ep esen a i eness: The di e si y o e ms and
emo ions was e i ied, con i ming ha he h ee se s
o e ed a balanced sample o ins i u ional, media, and
social na a i es. An analysis o geog aphical and
empo al co e age was included, which ensu ed ha
he da a e lec ed a ia ions a di e en s ages o he
ou b eaks.
In addi ion, calcula ions o a e age sen imen sco es and
con as es s we e added o he ex ual me ics. S uden 's - es
and ANOVA we e applied, yielding s a is ically signi ican
di e ences be ween sou ces in he dimensions o us , ea ,
and sadness. Con idence in e als o 95% we e epo ed,
ein o cing he alidi y o he compa isons.
The igu es we e gene a ed in English o main ain
consis ency. Wo d clouds, ba cha s, and sca e plo s we e
used. Each igu e included comple e desc ip ions o da a
sou ce, ime ange, and colo meaning. Colo g adien s
di e en ia ed emo ional in ensi y: da k ones indica ed
nega i e emo ions, while ligh ones ep esen ed posi i e o
neu al emo ions. This isual coding allowed o quick and
accu a e in e p e a ion o emo ional a ia ions.
IV. RESULTS
The da a analysis, ca ied ou in Janua y 2025, ocused on
he p ocessing and e alua ion o ex ual in o ma ion ex ac ed
om h ee main sou ces: o icial and al e na i e pla o ms.
This app oach combined ad anced ex mining ools,
sen imen analysis, and da a isualiza ion using wo d clouds
and ba cha s, allowing o a comp ehensi e in e p e a ion o
he emo ional and linguis ic na a i es p esen in each sou ce.
To inc ease igo , he emo ional analysis was no only
pe o med wi h Bing and NRC lexicons, bu also applied a
Nai e Bayes classi ie ained wi h manual anno a ions, which
allowed o he iden i ica ion o nuances beyond lexical
de ec ion. In addi ion, sen imen sco es we e no malized o a
ange o −1 o +1, and - es s and ANOVA wi h 95%
con idence in e als we e applied, con i ming di e ences
be ween sou ces.
The isualiza ions gene a ed, such as Figu e 2, used a
colo g adien ba designed o ep esen emo ions and hei
in ensi y. Blue ones we e associa ed wi h eelings o secu i y
and con idence, e lec ing posi i e emo ions ha p omo e
calm and op imism. In con as , eddish ones indica ed
nega i e emo ions such as dange , ea , and isk, allowing us
o iden i y a eas o g ea e ala m o conce n in he analyzed
ex s. This g adien is an e ec i e isual esou ce o
highligh ing emo ional ansi ions in he con en and has been
desc ibed in English in all igu es o main ain consis ency wi h
he language o he p ocessed e ms.
Wo d clouds we e used o highligh he mos equen
e ms in each da ase . In hese, he size o he wo ds e lec ed
hei p ominence wi hin he ex , while he in ensi y o he
colo ep esen ed he a e age pola i y calcula ed by he model.
Wa m ones such as o ange and ed indica ed wo ds wi h
nega i e conno a ions, while cool, bluish ones highligh ed
e ms associa ed wi h mo e posi i e o neu al emo ions.
These ools made i possible o iden i y speci ic pa e ns and
ends in he na a i es o each sou ce, helping o s uc u e he
esul s in a comp ehensible and e i iable manne . In addi ion,
i was obse ed ha he p esen a ion cha ac e is ics o each
pla o m (headlines and hema ic hie a chy in agg ega o s;
ins i u ional language and edi o ial guidelines in o icial
bodies; o es and mode a ion on Reddi ) shape he emo ional
exp ession o he con en and i s ecep ion by he public.
Figu e 2 is an illus a i e example o how colo s and shades
allow us o cap u e he emo ional ange con ained in he
analyzed ex s, acili a ing an in e p e a ion based on
s a is ical me ics.
Fig. 2. Colo g adien ba
A. Analysis o Da a om O icial Sou ces
O icial websi es p o ided eliable and s uc u ed
epidemiological da a, designed o communica e messages
ocused on p ac ical solu ions and conc e e measu es o
mi iga e he e ec s o epidemic ou b eaks. Sen imen analysis
showed a p edominance o posi i e emo ions, such as
con idence and an icipa ion, in line wi h hese ins i u ions'
goal o inspi ing calm, p omo ing in e na ional coope a ion,
and encou aging collabo a i e ac ion. Al hough nega i e
emo ions such as ea and sadness we e p esen , hey we e
add essed in a con olled manne , wi h an emphasis on
empa hy and e o s o o e come he challenges associa ed
wi h heal h c ises.
In addi ion o lexicons, a supe ised classi ie o emo ions
was applied, con i ming highe a e age sco es in posi i e
ca ego ies. Compa isons be ween alences we e e alua ed
wi h S uden 's - es , epo ing p < 0.05 and 95% CI no
including 0, which suppo s he di e ence obse ed in his
g oup. I should be no ed ha he que ies used we e
s anda dized in English (pandemic, con inemen , epidemic,
HMPV).
Figu e 3 illus a es how o icial websi es p io i ize
posi i e emo ions, such as us , o e nega i e emo ions,
which, al hough hey exis , a e p esen ed wi h less ala m. This
ep esen a ion ein o ces he op imis ic na u e o hei
na a i es, aimed a gene a ing secu i y and us in he public.
Fig. 3. Dis ibu ion o sen imen s on o icial pages.
On he o he hand, he wo d cloud gene a ed om o icial
publica ions highligh s e ms such as secu e, suppo ,
p o ec ion, and eme gency, associa ed wi h collec i e ac ion
and coope a ion be ween ins i u ions and ci izens. Figu e 4
shows his cloud: he size o each wo d co esponds o i s
equency o appea ance, and he p edominance o blue
indica es i s link o posi i e emo ions such as secu i y and
us .
In addi ion, he colo g adien o he cloud helps o
dis inguish he emo ional cha ge: mo e in ense blues indica e
wo ds linked o posi i e and ecu ing emo ions; ligh blues
highligh less equen e ms, al hough wi h op imis ic
conno a ions. Reds, p esen o a lesse ex en , ma k e ms
associa ed wi h ale o dange , which makes i easie o
ecognize nuances in he discou se o o icial sou ces.
Fig. 4. Wo d cloud gene a ed om o icial sou ces
B. Analysis o Da a om Google News
Analysis o da a ob ained om Google News showed a
p edominance o nega i e emo ions, such as sadness and ea ,
associa ed wi h he edi o ial endency o highligh ad e se and
high-impac aspec s. Ange was also obse ed, ela ed o
pe cep ions o social us a ion abou he managemen o he
heal h c isis and i s side e ec s. Posi i e emo ions, such as
con idence and joy, appea ed o a lesse ex en , indica ing less
a en ion o na a i es o p og ess o achie emen s. This
pa e n may ein o ce he pe cep ion o isk in audiences wi h
high exposu e o ala ming headlines.
S anda dized que ies in English (pandemic, con inemen ,
epidemic, HMPV) we e used in heal h/science ca ego ies.
Edi o ial bias owa d high-impac con en is ecognized,
a o ing a g ea e p esence o nega i ely cha ged e ms.
No malized sen imen sco es and ANOVA be ween g oups
showed a g ea e nega i e cha ge in his sou ce (p < 0.01; 95%
CI o he di e ence > 0).
Figu e 5 shows how nega i e emo ions a e mo e p e alen
in Google News con en . This pa e n coincides wi h media
na a i es ocused on se e i y and u gency.
Fig. 5. Dis ibu ion o sen imen s in Google News.
The wo d cloud gene a ed om hese pos s (Figu e 6)
suppo s his obse a ion, highligh ing e ms such as e e ,
dead, con ined, s ess, ou b eak, and dange ous. These wo ds
poin o isks, es ic ions, and un a o able consequences,
esul ing in an emo ionally cha ged na a i e. Te ms such as
wo s , p isone , and isk e lec expe iences wi h social and
psychological e ec s, while wo ds in da k ed ones, such as
dead and dange ous, ein o ce he p e ailing nega i e one. In
con as , hope and elie appea less equen ly and wi h less
isual p ominence.
The ch oma ic deg ada ion used in he cloud helps o
dis inguish he emo ions associa ed wi h he e ms analyzed.
In ense eddish ones indica e wo ds linked o nega i e
emo ions, while ligh e wa m ones e lec he low equency
o e ms wi h posi i e pe spec i es, such as p o ec ion o
solida i y. This allows us o see how he in ensi y o he colo
and he size o he wo ds con ey use ul in o ma ion abou he
emo ional cha ge and impo ance o he e ms in he con ex
analyzed.
Fig. 6. Wo d cloud gene a ed om Google News
C. Analysis o Da a om Reddi
Reddi was iden i ied as a pla o m whe e use s sha e
expe iences, opinions, and emo ions ela ed o epidemic
ou b eaks. The analysis was based on speci ic que ies in
sub eddi s such as /epidemiology, /heal h, and /wo ldnews.
These communi ies allow us o obse e pos s wi h di e se
pe spec i es on p e ailing conce ns and emo ions ela ed o
ou b eaks. The same que ies we e applied in English and
il e ed by language, emo ing URLs, emojis, and epe i ions
o educe noise. Communi y bias is ecognized due o he ules
and cul u e o each sub eddi , which can skew opics o ones.
Sen imen analysis showed ha ea was he mos
ecu en emo ion, associa ed wi h unce ain y abou he
impac o ou b eaks, mobili y es ic ions, and pe cei ed isks.
Sadness also ea u ed p ominen ly, e lec ing he social and
emo ional e ec s o c ises, such as isola ion and human loss.
A he same ime, emo ions o an icipa ion we e de ec ed,
linked o expec a ions abou medical ad ances, go e nmen
decisions, and possible solu ions.
The supe ised classi ie de ec ed a ec i e mix u es, and
S uden 's - es s be ween pola i ies yielded p < 0.05 wi h a
95% CI no including 0, suppo ing he p edominance o
nega i e emo ions o e posi i e ones in his sou ce. Figu e 7
shows he dis ibu ion o emo ions in Reddi pos s, wi h ea
and sadness p edomina ing. Posi i e emo ions, such as
con idence and joy, appea less equen ly, indica ing a ocus
on conce ns and challenges.
Fig. 7. Dis ibu ion o eelings in pos s ex ac ed om Reddi .
The wo d cloud gene a ed om he pos s (Figu e 8)
e eals p ominen e ms such as i us, ou b eak, in ec ions,
isk, and epidemic. These wo ds e lec conce ns abou he
magni ude o he ou b eaks and he se e i y o hei
consequences. Te ms such as bad and se e e ein o ce an
ala mis one in he discussions, while posi i e and secu e
appea less equen ly.
The size o he wo ds in he cloud e lec s hei equency
in he pos s: he la ges e ms, such as i us and isk, a e hose
ha appea mos equen ly. This isual ep esen a ion helps
o iden i y he mos discussed opics. In addi ion, he colo
deg ada ion applied allows he emo ional in ensi y o he
e ms o be dis inguished. Da k ones highligh wo ds
associa ed wi h nega i e emo ions, such as isk and se e e,
while wa m, ligh ones, p esen in posi i e, show a limi ed
p esence o op imis ic ideas. This isual app oach was
complemen ed by he supe ised classi ie , con i med by
es ing wi h IC 95%.
Fig. 8. Wo d cloud gene a ed om Reddi pos s
Reddi , he e o e, is iden i ied as a space whe e collec i e
discussions e lec social conce ns and, a he same ime,
a emp s a esilience in he ace o epidemic ou b eaks. The
use o size and colo s in he wo d cloud acili a es an in ui i e
g aphical ep esen a ion, showing how use s emo ionally
add ess he challenges associa ed wi h he pandemic.
D. Compa ison be ween o icial and al e na i e sou ces
The compa a i e analysis be ween o icial and al e na i e
sou ces highligh s ma ked di e ences in na a i es and
emo ions abou epidemic ou b eaks, showing how each g oup
communica es in o ma ion and how hese na a i es in luence
public pe cep ion.
O icial sou ces, ep esen ed in uchsia, ocused on
p omo ing messages o us , coope a ion, and secu i y. Te ms
such as secu e, websi es, o icial, and o ganiza ion domina ed
he na a i es, e lec ing language s uc u ed o gene a e calm.
The size o hese wo ds in he cloud illus a es hei
p ominence, while he ligh ones ein o ce he posi i e
emo ional cha ge. These ins i u ions p io i ized collabo a i e
ac ion and conc e e measu es o mi iga e he e ec s o he
ou b eaks. As shown in Figu e 9, hese guidelines seek o
s abilize public pe cep ion and p omo e in e na ional
coope a ion.
On he o he hand, Google News, iden i ied in pu ple,
p esen ed a mo e ala mis na a i e, ocusing on he impac o
he ou b eaks and hei isks. Te ms such as pandemic, isk,
con ined, and dange ous eme ge as he mos p ominen , wi h
da ke ones in ensi ying he nega i e conno a ion. This
app oach, e lec ed in Figu e 9, poin s o con en ha
highligh s se e i y and u gency, wi h po en ial e ec s on isk
pe cep ion and collec i e s ess.
Reddi , ep esen ed in g een, showed a mo e di e se and
pa icipa o y na a i e, wi h e ms o ala m and also o hope.
Wo ds such as i us, epidemic, and lock we e equen , while
hope and secu e e lec a emp s o egain op imism. The
g een ones, wi h g ada ions be ween ligh and da k, illus a e
he mix o emo ions, om ea o hope. As shown in Figu e 8,
Reddi unc ions as a space o collec i e exp ession whe e
emo ions and pe sonal expe iences coexis .
Figu e 9 summa izes hese di e ences in a compa a i e
isual diag am. O icial sou ces s and ou o hei op imis ic
app oach wi h ligh ones and secu i y-o ien ed e ms. Google
News p io i izes ala ming na a i es h ough da k ones ha
con ey u gency and g a i y. Reddi o e s a balance be ween
nega i e and posi i e e ms, wi h a wide ange o emo ions
and pe spec i es. This analysis allows us o see how each ype
o sou ce pa icipa es in he cons uc ion o pe cep ions du ing
heal h c ises.
Fig. 9. Compa ison be ween o icial and al e na i e sou ces
V. DISCUSSION
The esul s ob ained in his s udy show no able di e ences
in he ep esen a ion and pe cep ion o emo ions ela ed o
epidemic ou b eaks on pla o ms such as o icial and
al e na i e sou ces. Beyond he me ics, i can be obse ed
ha he edi o ial aming and dissemina ion logic o each
pla o m can ele a e o a enua e he pe cep ion o isk, wi h
p ac ical e ec s on public a en ion and willingness o ake
ac ion.
A. Reddi as a space o social exp ession
On Reddi , a p edominance o ea and sadness was
iden i ied, accompanied by an icipa ion. These emo ions
espond o he pa icipa o y na u e o he pla o m, whe e
use s sha e pe sonal expe iences and collec i e opinions
wi hou ins i u ional il e s. This bias owa d expe ien ial
accoun s ein o ces discussions o ien ed owa d ale ness and
conce n, so Reddi wo ks be e as a mood ba ome e han as
a e i iable in o ma ion channel o immedia e decisions.
B. Op imis ic na a i e in o icial sou ces
O icial sou ces we e cha ac e ized by mo e con olled
communica ion, ocusing on posi i e eelings such as
con idence and an icipa ion. These ins i u ions seek o con ey
calm and encou age coope a ion h ough messages designed
o a oid panic.[4] and [6] suppo his end, indica ing ha
in e na ional o ganiza ions end o p io i ize na a i es ha
p omo e secu i y in imes o c isis. Howe e , c i icism o his
s a egy was also iden i ied, as i can seem dis an om
indi idual expe iences, as men ioned in [19] . This disconnec
can lead o a pe cep ion o ins i u ional coldness in imes o
unce ain y.
C. Ala mis na a i e in Google News
Analysis o Google News publica ions showed a
p edominance o nega i e emo ions, such as ea , sadness, and
ange . This app oach emphasizes he mos ala ming aspec s o
he news, aligning wi h s udies such as [2], which indica e ha
he media o en p io i ize shocking con en o cap u e he
a en ion o hei audiences. Te ms such as “con ined,”
“dange ous,” and “ isk” a e ecu ing examples in his
na a i e. [15], documen s how his ype o app oach can
ampli y he public's pe cep ion o isk, gene a ing knock-on
e ec s. [11] wa ns ha his end can inc ease s ess in
audiences and make i di icul o make in o med decisions.
The wo d cloud gene a ed om Google News ein o ces
his obse a ion, showing a g ea e ep esen a ion o e ms
ela ed o isks and limi a ions. Al hough wo ds such as hope
and elie a e p esen , hei lowe equency e lec s a educed
in e es in highligh ing p og ess o posi i e elemen s. This
highligh s he need o balance media na a i es wi h messages
ha p omo e esilience and con idence.
D. Impac o echnologies and da a analysis
The use o ad anced echnologies such as APIs, ex
mining, and sen imen analysis was cen al o his s udy. Tools
such as es , h , ggplo 2, ex da a, and wo dcloud2
acili a ed bo h he analysis and isualiza ion o la ge olumes
o da a, allowing o he explo a ion o emo ional and
linguis ic pa e ns. The combina ion o hese me hods wi h
lexicons such as Bing and NRC made i possible o iden i y
p edominan emo ions and map how hey a e e lec ed in he
na a i es o di e en pla o ms.
In addi ion, hese echnologies o ganized he da a and
showed how a iables such as size, colo in ensi y, and shades
in wo d clouds help o in e p e he esul s isually. Fo
example, he use o g adien s made i possible o dis inguish
he emo ions associa ed wi h speci ic e ms, which acili a ed
he de ec ion o di e en ia ed na a i e and emo ional pa e ns
be ween o icial and al e na i e sou ces.
VI. CONCLUSIONS
The analysis showed ha o icial sou ces, such as hose
managed by he WHO, UN, and CDC, a e gea ed owa d
con eying con idence and p omo ing coope a ion among
audiences. These na a i es seek o minimize panic and
gene a e secu i y h ough s uc u ed messages. Al hough less
p e alen , nega i e emo ions also appea ed, associa ed wi h
he communica ion o isks and challenges, which we e
handled in a con olled manne o a oid unnecessa y ala m.
On he o he hand, al e na i e pla o ms such as Google
News showed a endency owa d ala mis na a i es ha
highligh elemen s o isk and unce ain y, which can inc ease
he pe cep ion o s ess in audiences. The p edominan
p esence o e ms linked o dange s and es ic ions shows an
edi o ial s yle ocused on cap u ing a en ion h ough high-
impac headlines.
G ea e neu ali y was expec ed on Reddi ; howe e , he
analysis e ealed a mix u e o ala mis accoun s wi h isola ed
a emp s a op imism. Discussions end o ocus on in ense
conce ns and a ange o nega i e emo ions, wi h occasional
appea ances o hope and esilience
These esul s should be iewed as p elimina y and
explo a o y. Based on hem, p ac ical applica ions a e
p oposed: eal- ime moni o ing dashboa ds ha in eg a e
media and social media signals, ea ly wa nings based on
pola i y shi s, and segmen ed communica ion campaigns ha
combine ins i u ional messages wi h eadings o he public's
emo ional s a e. Toge he , hey o e a comp ehensi e
o e iew o how he na a i es o each sou ce in luence
collec i e pe cep ion du ing heal h c ises.
BIBLIOGRAPHICAL REFERENCES
[1] K. H. Mangu i, R. N. Ramadhan, and P. R. Mohammed Amin,
“Twi e Sen imen Analysis on Wo ldwide COVID-19
Ou b eaks,” Ku dis an Jou nal o Applied Resea ch, pp. 54–65,
May 2020, doi: 10.24017/co id.8.
[2] A. Joshi, S. Ka imi, R. Spa ks, C. Pa is, and C. Raina Macin y e,
“Su ey o Tex -based Epidemic In elligence,” ACM Compu
Su , ol. 52, no. 6, No . 2019, doi: 10.1145/3361141.
[3] K. She a e al., “Cha ac e ising in o ma ion gains and losses
when collec ing mul iple epidemic model ou pu s,” Jou nal
Epidemics, ol. 47, Jun. 2024, doi:
10.1016/J.EPIDEM.2024.100765.
[4] J. A. Polonsky e al., “Ou b eak analy ics: a de eloping da a
science o in o ming he esponse o eme ging pa hogens,”
Philosophical T ansac ions o he Royal Socie y B: Biological
Sciences, 2019, doi: 10.1098/ s b.2018.0276.
[5] K. O. Bazile ych e al., “In o ma ion sys em o assessing he
in o ma i eness o an epidemic p ocess ea u e,” Sys em esea ch
and in o ma ion echnologies, ol. 2023, no. 4, pp. 100–112, Dec.
2023, doi: 10.20535/SRIT.2308-8893.2023.4.08.
[6] A. N. Desai e al., “Real- ime Epidemic Fo ecas ing: Challenges
and Oppo uni ies,” Heal h Secu , ol. 17, no. 4, p. 268, Jul. 2019,
doi: 10.1089/HS.2019.0022.
[7] J. L. He e a-Dies a, J. M. Buldú, M. Cha ez, and J. H. Ma ínez,
“Using symbolic ne wo ks o analyse dynamical p ope ies o
disease ou b eaks,” P oceedings o he Royal Socie y A:
Ma hema ical, Physical and Enginee ing Sciences, ol. 476, no.
2236, Ap . 2020, doi: 10.48550/a Xi .1911.05646
[8] T. H. Nguyen, M. Fisichella, and K. Rud a, “A T us wo hy
App oach o Classi y and Analyze Epidemic-Rela ed In o ma ion
F om Mic oblogs,” IEEE T ans Compu Soc Sys , 2024, doi:
10.1109/TCSS.2024.3391395.
[9] J. Tolles and T. Luong, “Modeling Epidemics Wi h
Compa men al Models,” Jou nal Jama Ne wo k, ol. 323, no. 24,
pp. 2515–2516, Jun. 2020, doi: 10.1001/JAMA.2020.8420.
[10] M. Ma ani, G. G. Ka ul, W. K. Pan, and A. J. Pa ola i, “In ensi y
and equency o ex eme no el epidemics,” P oceedings o he
Na ional Academy o Sciences, 2021, doi:
10.1073/pnas.2105482118/-/DCSupplemen al.
[11] A. B auns ein, L. Budzynski, and M. Ma iani, “S a is ical
mechanics o in e ence in epidemic sp eading,” Phys Re E, ol.
108, no. 6, Dec. 2023, doi: 10.1103/PhysRe E.108.064302.
[12] J. Wu, Z. Niu, and X. Liu, “Unde s anding epidemic sp ead
pa e ns: a isual analysis app oach,” Heal h Sys ems, ol. 13, no.
3, pp. 229–245, Jul. 2024, doi: 10.1080/20476965.2024.2308286.
[13] S. G acy, P. E. Pa e, H. Sandbe g, and K. H. Johansson, “Analysis
and dis ibu ed con ol o pe iodic epidemic p ocesses,” IEEE
T ans Con ol Ne w Sys , ol. 8, no. 1, pp. 123–134, Ma . 2021,
doi: 10.1109/TCNS.2020.3017717.
[14] K. M. A. Kabi and J. Tanimo o, “Analysis o epidemic ou b eaks
in wo-laye ne wo ks wi h di e en s uc u es o in o ma ion
sp eading and disease di usion,” Commun Nonlinea Sci Nume
Simul, ol. 72, pp. 565–574, Jun. 2019, doi:
10.1016/J.CNSNS.2019.01.020.
[15] Z. Wang, C. Xia, Z. Chen, and G. Chen, “Epidemic P opaga ion
wi h Posi i e and Nega i e P e en i e In o ma ion in Mul iplex
Ne wo ks,” IEEE T ans Cybe n, ol. 51, no. 3, pp. 1454–1462,
Ma . 2021, doi: 10.1109/TCYB.2019.2960605.
[16] B. Wang, M. Gou, and Y. Han, “Impac s o in o ma ion
p opaga ion on epidemic sp ead o e di e en mig a ion ou es,”
Nonlinea Dyn, ol. 105, no. 4, pp. 3835–3847, Sep. 2021, doi:
10.1007/S11071-021-06791-8/METRICS.
[17] Z. Wang, X. Rui, G. Yuan, J. Cui, and T. Hadzibegano ic,
“Endemic in o ma ion-con agion ou b eaks in complex ne wo ks
wi h po en ial sp eade s based ecu en -s a e ansmission
dynamics,” Physica A: S a is ical Mechanics and i s Applica ions,
ol. 573, Jul. 2021, doi: 10.1016/J.PHYSA.2021.125907.
[18] S. S. Chikka addi and G. R. Smi ha, “Epidemic Disease Expe
Sys em,” 1s IEEE In e na ional Con e ence on Ad ances in
In o ma ion Technology, ICAIT 2019 - P oceedings, pp. 571–576,
Jul. 2019, doi: 10.1109/ICAIT47043.2019.8987421.
[19] K. Osadcha, V. Osadchyi, and V. K uglyk, “The ole o
in o ma ion and communica ion echnologies in epidemics: an
a emp a analysis,” Uk ainian Jou nal o Educa ional S udies
and In o ma ion Technology, p., 2020, doi:
10.32919/uesi .2020.01.06.
[20] M. Imanipou , M. Shahma i, │ Saeideh, A. Mahkooyeh, A.
Ghobadi, and P. Sanja i, “Re lec ions on heal h in o ma ion
sou ces in epidemics in synch ony wi h he COVID-19 pandemic:
A scoping e iew,” Jou nal o Nu sing Ad ances in Clinical
Sciences, ol. 1, 2024, doi: 10.32598/JNACS.2401.1005.
[21] S. L. Peng e al., “NLSI: An inno a i e me hod o loca e epidemic
sou ces on he SEIR p opaga ion model,” An in e disciplina y
Jou nal o NonLinea Science, ol. 33, no. 8, Aug. 2023, doi:
10.1063/5.0152859.
