The Numbe s o Fea : Me hodological No e and Resul s
Agnese Va danega Claudia Va danega
Oc obe 1, 2025
Abs ac
This epo p esen s he me hodology and esul s o a quan i a i e con en analysis o I alian
media discou se du ing he i s COVID-19 lockdown (Feb ua y-May 2020). The s udy in es iga es
he hypo hesis ha he pandemic na a i e was ‘da ais ’—ma ked by a sepa a ion be ween nume ical
epo ing and scien i ic con ex —and ca ied an anxie y-inducing emo ional p o ile consis en wi h
pos - u h dynamics.
Based on a co pus o 2,144 news headlines, we employed LDA o iden i y hemes and sen imen
analysis o assess hei emo ional one. Ou indings e eal a clea hema ic di ide be ween a
“numbe s” clus e ( hemes Nume i,Bolle ini) and a “disciplina y” clus e (Scienza,Espe i,Misu e).
The “numbe s” clus e , in pa icula , is associa ed wi h a nega i e, high-a ousal emo ional p o ile.
These esul s suppo he ini ial hypo hesis ha he “da a ica ion” o he pandemic na a i e (da a
wi hou con ex ) unc ioned as a d i e o emo ional ac i a ion.
This documen accompanies he p esen a ion “I nume i della pau a: go e namen ali à pandemica
a opaci à e aspa enza” (“The Numbe s o Fea : Pandemic Go e nmen ali y be ween Opaci y and
T anspa ency”), p esen ed a he con e ence “The G ea Fea : Epidemics in he I alian Peninsula om
he 17 h Cen u y o Today,” Uni e si y o Te amo, Sep embe 30 - Oc obe 1, 2025. I s pu pose is o
make he da a discussed on ha occasion a ailable.
1 In oduc ion
The esul s o a con en analysis pe o med on he headlines o news ela ed o he co ona i us du ing
he i s lockdown pe iod a e p esen ed.
The i s pa o his s udy aimed o explo e he discou se su ounding he pandemic, as i was cons i u ed
in he i s phase o he co ona i us sp ead (Feb ua y 22 - May 15, 2020; §2), h ough he iden i ica ion
o hemes, ac o s, and ca ego ies.
The hypo hesis a he cen e o his epo is ha he pandemic da a we e p esen ed in a “da ais ”
manne (Van Dijck, 2014), and ha his concep ion o da a is closely linked wi h pos - u h (Shel on,
2020), unde s ood as a se o discu si e p ac ices and sha ed belie s (Fe a is, 2017;Lo usso, 2018).
The La en Di ichle Alloca ion (§3) and sen imen analysis p esen ed in his documen p o ided da a
o suppo he i s hypo hesis. The na a i e o he da a ( heme Nume i: Numbe s) was:
• clea ly sepa a ed om ha o Scienza (Science), Espe i (Expe s), and Misu e (go e nemen
Measu es) (§3.1);
• p e alen , and g ew o e ime, a he expense o o he hemes (§3.2), in a p ocess o da a ica ion
o he na a i e;
• cha ac e ized by an anxie y-inducing emo ional p o ile, while ins i u ional communica ion (Scienza,
Espe i, and Misu e) ollowed he s anda ds o eme gency communica ion (posi i e, calm, in con ol;
§4). The da a became a ool o emo ional ac i a ion a he han in o ma ion.
The las pa o he p esen a ion conce ned he quali a i e analysis o he coun e -na a i es p esen ed
in he a icles (and no jus in he headlines), and he use o da a o suppo hem.
1
2 The co pus: headlines on Google
We conside ed he headlines o newspape s and news si es ha included he wo d “co ona i us,” selec ing
hem om he op hi y Google sea ch esul s, in he pe iod be ween Feb ua y 22 and May 15, 2020.
This pe iod was chosen in conside a ion o he olume o use sea ches o he keywo d “co ona i us” on
Google (Fig. 1; sou ce: Google T ends). To educe geoloca ion (and indexing) bias, we chose a numbe
o esul s ha ended o p oduce duplica es o e he days, ensu ing a good a ie y o local sou ces.
The collec ed headlines we e p e-p ocessed and analyzed wi h he help o he R so wa e (R Co e Team,
2025)1.
The choice o ocus on he headlines o all sea ch esul s, a he han on majo newspape s, is mo i a ed
by wo conside a ions.
• The public o en limi s hemsel es o eading headlines, especially on social media.
• Mo eo e , while in he i s weeks o he pandemic (Ma ch 2020) he Audiweb ankings we e led
by Repubblica and Co ie e della Se a, ollowed by TgCom24 and Il Messagge o (Cazzola, 2020a),
hose on social in e ac ions show he ele ance o ee news ou le s like Fanpage.i ( hi d, p eceded
by Co ie e and ollowed in i h place by Repubblica: Cazzola, 2020c). The Comsco e anking o
he same pe iod is led by he publishe o Fanpage (Ciaopeople), while Il Co ie e della Se a is only
in six h posi ion (Cazzola, 2020b).
While he sample canno be said o be ep esen a i e o wha was published by news si es, i is indica i e
o he ype o news ha had he g ea es ci cula ion in he conside ed pe iod: a o al o 2,144 headlines2,
o 248 online ou le s (na ional, local, ele ision; Table 1).
Figu e 1: Web sea ches o he e ms ”co ona i us” and ”co id”
Da es ep esen ed in he g aph. Feb ua y 23: Es ablishmen o he i s ed zone (peak o sea ches);
Ma ch 9: The “escape om he No h”; Ma ch 11: “Cu a I alia” dec ee (“I s ay a home”); Ap il 4:
manda o y masks; Ap il 18: pos ponemen o he easing o es ic i e measu es; May 16: Phase 2 Dec ee.
1In pa icula , he Quan eda (Benoi e al., 2025) and opicmodels (G ün & Ho nik, 2024; see also Phan e al., 2008)
2
Table 1: Main news ou le s included in he sample
News ou le N
co ie e.i 130
epubblica.i 117
las ampa.i 101
ilsole24o e.com 88
ilmessagge o.i 87
ansa.i 86
il a oquo idiano.i 82
gcom24.mediase .i 69
ainews.i 63
open.online 53
adnk onos.com 52
lanazione.i 41
g24.sky.i 41
il es odelca lino.i 39
agi.i 37
quo idiano.ne 37
ilpos .i 35
geno a24.i 33
ilgio no.i 33
wi ed.i 33
i g.i 30
ilci adinomb.i 29
lanuo asa degna.i 22
ilma ino.i 18
in e nazionale.i 17
lagazze adelmezzogio no.i 17
libe oquo idiano.i 17
lap o inciac .i 16
la ocedel en ino.i 16
al alex.com 15
o miche.ne 15
il empo.i 15
ilgiunco.ne 14
qui inanza.i 14
spo .sky.i 14
anpage.i 13
il i eno.gelocal.i 13
co ie edellospo .i 12
lap essa.i 12
ocus.i 11
la7.i 11
be gamonews.i 10
ilcapoluogo.i 10
il oglio.i 10
ilsecoloxix.i 10
i ie a24.i 10
salu e.go .i 10
il a oalimen a e.i 9
m.c onachemace a esi.i 9
medical ac s.i 9
News ou le N
picchionews.i 9
ani y ai .i 9
i a.i 9
ecodibe gamo.i 8
co ie ecomunicazioni.i 7
o bes.i 7
ilcen o.i 7
luccaindi e a.i 7
a esenews.i 7
co ie edicomo.i 6
in oda a.ilsole24o e.com 6
ispionline.i 6
money.i 6
na ionalgeog aphic.i 6
a a i aliani.i 5
askanews.i 5
a eni e.i 5
chie i oday.i 5
co ie ead ia ico.i 5
co ie e omagna.i 5
c emonaoggi.i 5
ondazione e onesi.i 5
gazze a.i 5
ildolomi i.i 5
ilgazze ino.i 5
ilme eo.i 5
ilpesca a.i 5
i .businessinside .com 5
lasiciliaweb.i 5
milano oday.i 5
mo o i. i gilio.i 5
nonsp eca e.i 5
e e8.i 5
oma oday.i 5
a icannews. a 5
bologna oday.i 4
co ie edellumb ia.co .i 4
ilpiccolo.ne 4
iodonna.i 4
osse a o iomala ie a e.i 4
a ennano izie.i 4
sulpana o.ne 4
3
3 Topic modeling: LDA
The opic modeling analysis was conduc ed on a ma ix de ined by 625 “cha ac e izing” e ms, i.e., wi h
high ela i e equency bu p esen in a limi ed numbe o ex s (10%). Following he e m educ ion,
he numbe o analyzable headlines became 2,102.
To de e mine he numbe o opics o ex ac and he dis ibu ion’s 𝛼pa ame e (which egula es he
numbe o opics a ibu ed o each documen ), we es ed se e al models3. Finally, a 7- opic model wi h
𝛼equal o 0.2 was chosen4.
Figu e 2: LDA. Composi ion o he hemes: ”numbe s” clus e
In LDA (D. Blei e al., 2001;D. M. Blei e al., 2003), a opic is de ined as a la en dimension ha
o ganizes he ocabula y o a co pus. The p ocedu e econs uc s a pos e io dis ibu ion o e ms o
each opic (𝛽ma ix), and one o opics o each headline (𝛾ma ix)5. This is a echnique we could call
induc i e, hus sui able o he ype o sample used, o explo a o y pu poses.
Seman ic in e p e a ion o he hemes This is achie ed by using he in o ma ion in he 𝛽ma ix
( e ms- opics). Figs. 2and 3p esen he 10 mos signi ican e ms o each heme ( he measu e is no
indica i e o equency).
Fo he pu pose o he compa ison ele an he e, i e o he se en hemes a e di ided in o wo main
mac o-clus e s (§3.1):
packages.
22,555 a e emo ing duplica es, which become 2,144 a e elimina ing hose ou side he da e ange o no classi iable
as “news.”
3Wi h he lda uning package ( o calcula ing he op imal numbe o clus e s: Niki a, 2020) and opicdoc ( o a pos e io i
quali y measu es: F iedman, 2022).
4To educe he opics assignable o each documen , compa ed o he de aul pa ame e (0.1), gi en he b e i y o he
ex s.
5Fo he applica ion o his echnique o e y sho ex s, c . Albalawi e al. (2020).
4
• a disciplina y clus e (cool colo s): Scienza (Science), Espe i (Expe s), and Misu e (Measu es)
and
• a numbe s clus e (wa m colo s): Nume i (Numbe s) and Bolle ini (Bulle ins).6
The C onaca (Daily News) heme la gely ollows he beha io o he la e , while he Es e i (Wo ld)
heme is qui e a ied in i s composi ion.
Figu e 3: LDA. Composi ion o he hemes: ”disciplina y” clus e
E alua ion o he esul s Fo he e alua ion o he hemes, a se ies o in o ma ion was conside ed,
also ele an o ou discou se, and in pa icula o show ha :
• he mos widesp ead hemes a e Nume i and Scienza;
• he Nume i and Bolle ini hemes a e no e y a ied and no e y in o ma i e, in he sense ha
hey con ain a he common e ms (which do no s and ou om he co pus);
• while, on he con a y, he Scienza heme is he mos in o ma i e and seman ically a ied.
The measu es, p esen ed in Table 2, can be ead as ollows7:
•P ominence: numbe o documen s in which a heme is p esen ( ega dless o i s weigh ).
•F equency: numbe o documen s uniquely assigned o he heme (based on he highes alue).
•Size: he numbe o e ms o ac ions o e ms p esen in he dis ibu ion o each heme. The
o al is 625, he numbe o e ms in he ma ix.
6Theme glossa y: Scienza = Science; Espe i = Expe s; Misu e = (go e nmen ) Measu es; Nume i = Numbe s;
Bolle ini = (Ci il P o ec ion) Bulle ins; C onaca = Daily News; Es e i = Wo ld News.
7O he a ailable measu es, such as he dis ance in e ms o -id and he cohe ence index, which e e o he dis ibu ions
o e ms in he ex s, a e no app op ia e in his case, gi en he b e i y o he ex s hemsel es ( o de ails on he calcula ion
and use o he measu es, c . Ai oldi e al., 2015, p. 238 e seq.).
5
Table 2: Theme cha ac e is ics
Theme P ominence
(no. headlines) F equency Size
( e ms)
Exclusi i y
( e ms)
Dis ance
om co pus
Misu e 436 214 109,52 9,68 0,60
Es e i 422 201 94,44 9,64 0,58
C onaca 461 212 89,32 9,69 0,57
Espe i 446 204 104,40 9,47 0,60
Scienza 465 268 113,13 9,58 0,61
Nume i 539 306 67,27 9,66 0,55
Bolle ini 416 226 46,91 9,86 0,59
To al 2.102 1.631 625,00 10,00 –
–The mos seman ically a ied heme is Scienza;
–The leas di e en ia ed is Bolle ini, p eceded by Nume i (a e all, numbe s a e no included
in he e m-documen ma ix).
•Exclusi i y: how many o he mos impo an e ms ( he op en, o example) a e exclusi e o
he heme i sel . Conside ing he echnique adop ed, he b e i y o he ex s, and he chosen 𝛼
alue, we ha e a s ong speci ici y o he hemes (see also Fig. 5). This means ha he hemes a e
well-dis inc om each o he .
•Dis ance (Hellinge ) om he co pus: di e gence be ween he dis ibu ion o e ms in he heme
and ha in he en i e co pus. The g ea e he dis ance, he mo e in o ma i e he iden i ied heme
is:
– he mos dis an /in o ma i e is Scienza.
– he leas dis an is Nume i.
Figu e 4: LDA. Dend og am o he hemes
3.1 The sepa a ion o “numbe s” and science
While he wo mos equen and ans e sal hemes, Nume i and Scienza ep esen comple ely dis inc
“na a i es” o he pandemic.
6
The dend og am in Fig. 4highligh s ha he numbe s mac o-clus e (Nume i and Bolle ini) sepa a es
clea ly and immedia ely om he o he hemes, pa icula ly om Scienza and Espe i.
This also eme ges in mo e de ail in he g aph in Fig. 5, which ep esen s he connec ions be ween
hemes and e ms (i.e., he 𝛽ma ix). The p esence o connec ions (sha ed e ms) be ween hemes can
be in e p e ed as indica i e o seman ic p oximi y.
Figu e 5: LDA. Theme-Te m G aph (be a)
The able in Fig. 6p o ides u he con i ma ion o his sepa a ion.
The alues ep esen ed a e he means o he sco es om he 𝛾(gamma) ma ix, which is he p obabili y
ha a headline is associa ed wi h a heme ( he h eshold is 0.14, i.e., 1 / 7 hemes): he s ong p esence
o numbe s (3 o mo e occu ences wi hin he headline) cha ac e izes he Nume i and Bolle ini clus e s,
while Scienza and Espe i a e cha ac e ized by hei absence8.
Figu e 6: Mean gamma alues o headlines wi h numbe s
8The numbe s a iable ep esen s he coun o numbe s in he headlines, in wo ds o digi s. These okens we e iden i ied
ia POS agging and hen manually co ec ed, o exclude da es and numbe s con ained in he exp essions such as “co id
19”, “sa s-co 2”, e c. Numbe s in digi s did no con ibu e o he de ini ion o he hemes.
7
3.2 Numbe s p e ail o e ime
The a e age daily dis ibu ion o hemes shows how hey de elop o e ime (Fig. 79): he Nume i heme
inc easingly accoun s o a la ge sha e o he discou se, a he expense, in pa icula , o Scienza,Espe i,
and Misu e.
Figu e 7: Themes o e ime (p esence; 15-day mo ing a e age)
4 Sen imen analysis
The sen imen analysis was ca ied ou wi h he a ec i e dic iona y ELI a (Di Palma, 2024b,2024a),
which p o ides sco es o 6,905 I alian lexical o ms, on wo classi ica ions o emo ions: he VAD model
(§4.1) and Plu chik’s wheel o emo ions (§4.2).
The agg ega ed sco es o he headlines by heme con i m he dis ance be ween he wo mac o-clus e s,
which ha e opposi e emo ional p o iles. This esul is all he mo e signi ican as:
• i elies on wo di e en ypes o classi ica ion o he a ec i e one o he ex s (no simply
posi i e/nega i e);
• a heme is a ibu ed by he minimum h eshold o he 𝛾ma ix alue, and his ends o la en
he esul s (each documen can belong o mul iple hemes).
4.1 The VAD model
The VAD model (Russell, 1980;Russell & Meh abian, 1977), a co ne s one o he dimensional analysis
o emo ions, is a icula ed on h ee dimensions:
9Mo ing a e age o he daily 𝛾 alue; he o al o each day is one; he signi icance h eshold is 0.14.
8
Table 3: Mean VAD sco es by heme
Theme Valence A ousal Dominance
Bolle ini 0,150 0,534 -0,091
C onaca 0,052 0,549 -0,190
Espe i 0,223 0,448 -0,012
Es e i 0,105 0,534 -0,105
Misu e 0,243 0,453 -0,023
Nume i -0,105 0,560 -0,325
Scienza 0,184 0,504 -0,045
– 0,122 0,512 -0,113
Scale: -4, +4; gamma min.: 0.14
•Valence. A ec i e pola i y (unpleasan /pleasan ).
•A ousal. Physiological and psychological in ensi y o an emo ion. A mo e a ousing emo ion is
mo e in ense, ega dless o i s alence.
•Dominance (con ol). The pe cei ed sense o con ol o e a gi en emo ion o si ua ion10.
The a e age sco es, as men ioned abo e, a e e y low — close o ze o (neu ali y: Table 3). Howe e ,
he dis ance be ween he wo mac o-clus e s is e iden when conside ing he p o iles, de ined by he
di e ences om he means (Figs. 8and 9).
Figu e 8: VAD: numbe s clus e
10They closely esemble he seman ic opposi ions iden i ied as p ima y by Osgood e al. (1956): good-bad (e alua ion,
simila o alence), s ong/weak (ac i i y, a ousal) and ac i e/passi e (powe , dominance).
9
