Personality trait analysis during the COVID-19 pandemic: a comparative study on social media

Jou nal o In elligen In o ma ion Sys ems h ps://doi.o g/10.1007/s10844-023-00810-3 RESEARCH Pe sonali y ai analysis du ing he COVID-19 pandemic: a compa a i e s udy on social media Ma cos Fe nández-Pichel1·Ma io Ez a A agón1·Julián Sabo ido-Pa iño1· Da id E. Losada1 Recei ed: 28 June 2023 / Re ised: 16 Augus 2023 / Accep ed: 17 Augus 2023 © The Au ho (s) 2023 Abs ac The COVID-19 pandemic, a global con agion o co ona i us in ec ion caused by Se e e Acu e Respi a o y Synd ome Co ona i us 2 (SARS-CoV-2), has igge ed se e e social and economic dis up ion a ound he wo ld and p o oked changes in people’s beha io . Gi en he ex eme socie al impac o COVID-19, i becomes c ucial o unde s and he emo ional esponse o he people and he impac o COVID-19 on pe sonali y ai s and psychological dimensions. In his s udy, we con ibu e o his goal by ho oughly analyzing he e olu ion o pe sonali y and psychological aspec s in a la ge-scale collec ion o wee s ex ac ed du ing he COVID-19 pandemic. The objec i es o his esea ch a e: i) o p o ide e idence ha helps o unde s and he es ima ed impac o he pandemic on people’s empe amen , ii) o ind asso- cia ions and ends be ween speci ic e en s (e.g., s ages o ha sh con inemen ) and people’s eac ions, and iii) o s udy he e olu ion o mul iple pe sonali y aspec s, such as he deg ee o in o e sion o he le el o neu o icism. We also examine he de elopmen o emo ions, as a na u al complemen o he au oma ic analysis o he pe sonali y dimensions. To achie e ou goals, we ha e c ea ed wo la ge collec ions o wee s (geo agged in he Uni ed S a es and Spain, espec i ely), collec ed du ing he pandemic. Ou wo k e eals in e es ing ends in pe sonali y dimensions, emo ions, and e en s. Fo example, du ing he pandemic pe iod, we ound inc easing aces o in o e sion and neu o icism. Ano he in e es ing insigh om ou s udy is ha he mos equen signs o pe sonali y diso de s a e hose ela ed o dep es- sion, schizoph enia, and na cissism. We also ound some peaks o nega i e/posi i e emo ions ela ed o speci ic e en s. Keywo ds COVID-19 ·Social media ·Big-5 ·Pe sonali y analysis ·Emo ion analysis 1 In oduc ion The Co ona i us disease ou b eak (COVID-19) began in Decembe 2019 and apidly sp ead wo ldwide, causing millions o dea hs (Zhu e al., 2020). Mos people in ec ed wi h he i us BMa cos Fe nández-Pichel [email p o ec ed] 1Cen o Singula de In es igación en Tecnoloxías In elixen es (CiTIUS), Uni e si y o San iago de Compos ela, Rúa Jena o de la Fuen e, San iago de Compos ela 15782, A Co uña, Spain 123 Jou nal o In elligen In o ma ion Sys ems expe ienced mild o mode a e espi a o y illness and eco e ed wi hou equi ing special ea men . Howe e , some indi iduals de eloped se ious symp oms and equi ed medical a en ion (O ganiza ion, 2023). Olde people and hose wi h unde lying medical condi ions we e mo e likely o de elop se e e illnesses. COVID-19 has caused a huge impac wo ldwide on heal h sys ems, economies, and educa ion sys ems (O ganiza ion, 2020). A his poin , we s ill do no know how long his damage will las o when he wo ld will ully eco e . The pandemic had a majo e ec on ou li es, and many people ha e been acing s ess ul and o e whelming challenges. These s ess ac o s can cause changes in appe i e, ene gy, desi es, and in e es s, and can c ea e eelings o ea , ange , sadness, wo y, numbness, o us a ion. Lea ning o iden i y hese symp oms can help us o cope and manage heal hily. The inc easing social media ac i i y p esen s an oppo uni y o s udy he e olu ion o beha io al and pe sonali y ai s in a la ge p opo ion o he wo ld popula ion. Fo many people, social li e does no only happen in hei su oundings o immedia e en i onmen . In many cases, a conside able numbe o social in e ac ions ake place in i ual se ings c ea ed by social media pla o ms like Facebook, Twi e , o Reddi . Since he beginning o he pandemic, social media pla o ms a e inc easingly being used as an in o ma ion sou ce (e.g., o be in o med abou isks and c ises) and publica ion channel (Reu e s, 2022). Men al heal h has been a majo global conce n wi hin he las decades, bu psychological p oblems ha e agg a a ed since he con inemen pe iod o he yea 2020 (Gup a e al., 2020). Ha ing his in mind, he main mo i a ion o his wo k lies in he necessi y o u he s udy he g ow h o his “psychological pandemic” and o aise public awa eness. Fo example, by de eloping new sc eening ools able o in o m au ho i ies and he popula ion abou he impac o he pandemic on pe sonali y dimensions. I is also impo an o ma ch psychological ai s wi h speci ic e en s o ci cums ances ha happened du ing his di icul pe iod (e.g., ela ed o ha sh con inemen s). In addi ion, analyzing he e olu ion o eelings and emo ions is a na u al complemen o he s udy o pe sonali y dimensions. We will ocus on wo collec ions o wee s e ie ed du ing he wo s pa o he pan- demic pe iod (2020-2021). The wee s come om wo coun ies (USA and Spain), and we es ima e he p e alence and e olu ion o speci ic pe sonali y ai s and diso de s on hese wo la ge samples o social media en ies (10,234,223 and 17,395,598 wee s, espec i ely). To ha end, we d aw om ecognized psychological ins umen s, such as he Big-5, which de ines i e co e pe sonali y ac o s (ex a e sion, ag eeableness, openness o expe ience, conscien iousness, and neu o icism) (Ba ick and Moun , 1991). We also epo on ho opics ound in he published con en , wi h a pa icula ocus on hemes ha ha e igge ed con- ce ns, pola ized emo ions, o ex eme eelings. We can summa ize ou main con ibu ions as ollows: 1. Two publicly a ailable da ase s o wee s (abou gene al opics) ex ac ed du ing he mos p oblema ic mon hs o he COVID-19 pandemic. The collec ions con ain wee s w i en in English and Spanish and he wee s a e geo agged in USA o Spain, espec i ely. We also make a ailable o he communi y a eposi o y o acili a e he ep oducibili y o he expe imen s p esen ed in his s udy.1 2. A ho ough analysis o he p e alence and e olu ion o pe sonali y ai s in Twi e use s om wo di e en coun ies (USA and Spain). 3. As udyo hee olu iono signso pe sonali ydiso de s(Schizoid,Dep essi e,A oidan , Dependen , His ionic, Na cissis ic, Compulsi e, Pa anoid, and Schizo ypal) du ing he pandemic. 1h ps://gi hub.com/Ma cosFP97/COVID-19-Pe sonali y 123 Jou nal o In elligen In o ma ion Sys ems 4. A complemen a y epo o emo ions es ima ed om he wee s published du ing he pandemic. The emainde o he pape is o ganized as ollows. Sec ion 2, p esen s a b ie o e iew o he ela ed wo k. Sec ion 3, desc ibes in de ail ou me hodology o c ea e he da a and ou app oaches o analyze he publica ions. Sec ion 4 epo s he esul s and Sec ion 5p esen s a discussion o he ob ained esul s and limi a ions o his wo k. Finally, Sec ion 6gi es some concluding ema ks. 2 Rela ed wo k In his sec ion, we discuss some ecen wo ks ela ed o social media analysis du ing he pandemic pe iod. Di e en s udies ha e pe o med esea ch on he impac o COVID-19 on people’s public and pe sonal li es. One o he mos popula s a egies consis s o mining social media da a and measu ing conce ns o sen imen s (e.g., h ough publica ions pos ed by people on hei social media accoun s). In Lyu e al. (2020), he au ho s sea ched o wee s ha con ained e ms e e ing o he COVID-19 pandemic, such as “Chinese i us” o “Wuhan i us”, and hen compa ed hem wi h wee s ha did no ha e hese wo ds. This s udy e ealed di e ences in age, geoloca ion, and poli ical iews among he indi iduals who pos ed hese wo ypes o wee s. Voh a and Ga g (2023) pe o med sen imen analysis on Twi e da a con aining keywo ds ela ed o wo king om home. A o al numbe o 358,823 wee s we e collec ed. The au ho s i s labeled a subse o wee s using VADER and, nex , ained a Con olu ional Neu al Ne wo k (CNN) o es ima e sen imen . They ound ha he majo i y o pos s had a posi i e sen imen owa ds wo king om home. An in e es ing s udy ha analyzes popula ion-le el dispa i ies du ing he pandemic was p esen ed in Zhang e al. (2021). The au ho s i s collec ed a la ge-scale da ase om Twi e . Then, hey di ided he da a among popula ion g oups and ex ac ed conce ns, sen imen s, and emo ions. This wo k e ealed di e ences ega ding COVID-19 opics among popula ion g oups ( o example, gende , and age). One o he indings was ha he popula ion g oup composed o abo e 40-yea -old women was he g oup mos conce ned abou COVID-19. This segmen o emales was mos conce ned abou economics and educa ion, while males in he same age g oup we e mos conce ned abou poli ics and economics. In ano he s udy, Alhuzali e al. (2022) collec ed wee s ela ed o COVID-19 om 48 di e en ci ies in he Uni ed Kingdom. The sen imen , emo ion, and opics o he wee s we e examined using deep lea ning models like Sen icNe 6 and SpanEmo, wi h a combined opic modeling app oach. The au ho s ound ha people’s a i udes and exp essions we e highly posi i e a he beginning o 2020, bu s a ed o dec ease o e ime owa d he end o 2021. This app oach can po en ially supply aluable cues abou how public policies a e pe cei ed by people in di e en geog aphical a eas. In a sligh ly di e en di ec ion, Umai and Mascia i (2023) explo ed he sen imen s o he ci izens ela ed o he COVID-19 accina ion. This s udy pe o med sen imen and spa ial analyses o ex ual pieces o e idence. The au ho s used he Tex Blob ool o es ima e he pola i yo he ex sandca ego ized hem.Theyalsomapped heda aob ained ogeog aphical loca ions o ge a global pic u e o people’s a i udes owa d accina ion. Simila ly, in Bo ah (2023) a mul imodal deep lea ning me hod o Indian wee s classi ica ion was in oduced. The goal was o es ima e popula ion’s hesi ancy owa ds accina ion h ough social media pos s. The s udy e ealed ha con idence in accina ion inc eased wi h ime. 123 Jou nal o In elligen In o ma ion Sys ems Si e al. (2021) su eyed pos - auma ic s ess synd ome (PTSS) and sleep quali y among esiden s in Wuhan and nea by ci ies. To ha end, a PTSS checklis o DSM-5 (Diagnos ic and S a is ical Manual o Men al Diso de s) and ou ools om he Pi sbu gh Sleep Quali y Index (PSQI) we e u ilized. The su ey e ealed a p e alence o PTSS a e he COVID-19 ou b eak and showed ha women su e ed mo e han men. Mo e speci ically, an ad e se al e a ion in cogni ion and sleep quali y was associa ed wi h emales. The au ho s also ound ha pa icipan s wi h be e sleep quali y epo ed lowe PTSS. In Ahmad and Mu ad (2020), s udied how social media a ec s men al heal h and he sp ead o panic abou COVID-19 in he Ku dis an Region o I aq. They ound ha social media has a signi ican impac on sp eading ea and panic, and in luences nega i ely people’s men al heal h. Young people, aged be ween 18 and 35, we e ound o be acing psychological anxie y. Resea che s in Se e al. (2022) in es iga ed he opic e olu ion o Twi e o he Republic o Tu key in 2020. They analyzed 1.3 million wee s ela ed o he co ona i us be ween Feb ua y 24, 2020, and May 2, 2020. The au ho s ound an inc ease in opics ela ed o hygiene, li es yle, and anxie y. Inspi ed by he Wo ld Bank’s Po e y Moni o ing Technical No e, hey also examined he e ec o income on con en sha ing. Acco ding o he au ho s’ es ima es, use s wi h a lowe g oss domes ic p oduc pe capi a end o sha e mo e news ela ed o COVID-19 con en . In Ainley e al. (2021), he eac ion o people o heal h and ca e deli e y in he Uni ed Kingdom was s udied. The au ho s iden i ied he ollowing main hemes: access o emo e ca e, quali y o emo e ca e, he an icipa ion o emo e ca e, online booking, asynch onous communica ion, and publicizing changes o se ices o ca e deli e y. A he beginning o he pandemic, he commen s we e posi i e bu his posi i e end dec eased o e ime wi h he inc ease in es ic ions and COVID cases. In he a ea o pe sonali y analysis, we can also ind app oaches o le e age da a ha is sha ed on he In e ne on a daily basis. Fo example, deep lea ning app oaches show p omise in pe sonali y classi ica ion when an ample amoun o da a is a ailable o aining (Leona di e al., 2020;Die al.,2018). Howe e , hei limi ed explana o y unde s anding capabili ies hinde specialis s om comp ehending he a ionale behind au oma ic p edic ions. To y o sol e his, de-la Rosa e al. (2023) aimed o analyze language and disce n pe sonali y ai s om sho communica ions by ha nessing in o ma ion ex ac ed om a men al lexicon. The objec i e was o de elop a aluable ep esen a ion ha aids specialis s in unde s anding he key a iables ha con ibu e o he analysis o an indi idual’s pe sonali y. This highligh s he impo ance o explainabili y in his ype o ask. As we can see abo e, o me s udies on social media du ing he pandemic ha e mainly ocused on sen imen and emo ion analysis. These wo ks ha e essen ially explo ed he con- ce ns ha people aise and hei impo ance. Following his line o hough , we expand his ype o analysis by s udying pe sonali y ai s and pe sonali y diso de s. We es ima e he p e alence and e olu ion o pe sonali y ai s and diso de s du ing he pandemic using Deep Linguis ic echnologies. Ou wo k also di e s om he p e ious s udies by explici ly ela ing he esul s ob ained o speci ic social e en s. To ha end, we pe o m ou analysis on a collec- ion o wee s whose size is subs an ially la ge han he sizes conside ed by p e ious wo ks. Mo eo e , he da a is geo agged in wo di e en coun ies and he c awling/ex ac ion p o- cess was no biased owa ds speci ic keywo ds. We pe cei e he b oad scope o his esou ce as a s eng h, as i p o ides o he esea ch eams wi h he oppo uni y o ca y ou a a ie y o analyses. I also allows us o s udy cul u al di e ences and eac ions when handling a p oblem o he magni ude o COVID-19. The collec ion and he code a e a ailable o he communi y o es and euse. 123 Jou nal o In elligen In o ma ion Sys ems Fig. 1 Gene al iew o he sys em. We ex ac ed Twi e da a using Twin (sc apping ool) and, nex , we pe o med mul iple epo s wi h complemen a y analy ical ools 3 Me hodology In his sec ion, we p esen in de ail he s a egy implemen ed o collec ing wee s and we desc ibe ou me hods o analyzing pe sonali y ai s, diso de s, and emo ions. Figu e 1 depic s an o e all iew o all he s eps in ol ed. 3.1 Da a collec ion Fo collec ing da a, we employed Twin .2Twin o e s he possibili y o ex ac ing wee s in a geolocalized manne . The p ocess consis s o choosing a cen al geog aphical poin , es ablishing a adius, and ob aining wee s om ha a ea. Fo ou s udy, he poin s cho- sen we e Mad id and New Yo k ( o ep esen ing Spain and USA, espec i ely). A e he ex ac ion, wee s we e classi ied by language,3and we only kep he English wee s om he USA da ase and he Spanish wee s om he Spain da ase . Fo he da a p ocessing, we pe o med a simple p e-p ocessing o he ex s by emo ing special cha ac e s like URLs, @, and hash ags. The USA da ase con ains wee s whose publica ion da es ange om Janua y 2020 o Augus 2021. The o al numbe o wee s is 10,234,223. The Spain da ase con ains 17,395,598 wee s (publica ion da es om Janua y 2020 o May 2021). A subse o his da a is eely a ailable o he communi y o es and use.4We ha e ca e ully espec ed Twi e ’s policy ega ding anonymiza ion and he maximum numbe o pos s. Mo e speci ically, a he ime o c ea ing his collec ion, he limi s o Twi e es ablished ha “you may only dis ibu e up o 1,500,000 Twee IDs o a single en i y wi hin a 30-day pe iod”.5Those eade s in e es ed in he en i e da ase can con ac he au ho s, since his is a gene ic da ase ha could be exploi ed o o he esea ch pu poses. I is wo h men ioning ha we had some di icul ies du ing he cou se o he ex ac ion s age. The Twin se ice was shu down du ing he de elopmen o his p ojec . This a ec ed he e ie al o Spanish wee s. Mo e speci ically, o Spain’s da ase , he numbe o collec ed wee s du ing he cen al pe iod o he pandemic (a ound 4,561 wee s pe week) a e less han 2h ps://gi hub.com/ win p ojec / win 3To ha end, we employed h ps://pypi.o g/p ojec /langde ec /. 4h ps://gi hub.com/Ma cosFP97/COVID-19-Pe sonali y/ ee/mas e /da ase 5h ps://de elope . wi e .com/en/de elope - e ms/mo e-on- es ic ed-use-cases 123 Jou nal o In elligen In o ma ion Sys ems he numbe o collec ed wee s du ing he es o he imeline (a ound 345,849 wee s/pe week o Spain and 159,948 wee s/pe week o he USA). This should be aken in o accoun o he analysis, as ends du ing his pe iod show o en a la ge a iance. S ill, he o e all collec ion o wee s is subs an ially la ge han hose employed in mos exis ing s udies. 3.2 Big-5 pe sonali y analysis The main objec i e o his wo k is o unde s and he pe sonali y ai s and hei e olu ion in Social Media use s du ing he COVID pandemic. Pe sonali y has been de ined (Funde , 1997) as he “indi idual’s cha ac e is ic pa e ns o hough , emo ion, and beha io , oge he wi h he psychological mechanism behind hose pa e ns”. Howe e , his de ini ion is oo gene al and makes unde s anding indi idual di e ences in beha io and expe ience oo di icul . In he a ea o Psychology, esea ch on pe sonali y, led by mul iple eams o esea che s, eached a consensus on a gene al axonomy o pe sonali y ai s, named “Big Fi e” (John and S i as a a, 1997). Pe sonali y analysis (Ba ick and Moun , 1991) using Big-5 builds om a axonomy o classi ica ion p oposed in he 1980s. This model was de ined using ac o analysis (Yong and Pea ce, 2013), and s udies he ela ionship be ween a la ge numbe o e bal desc ip o s and pe sonali y ai s. The o iginal Big-5 model was p oposed by Digman, and la e ex ended by Goldbe g (1993). These de ini ions ha e been ound o con ain he majo i y o pe sonali y ai s and es ablish a solid s uc u e unde which pe sonali y can be o mally s udied. The i e ac o s iden i ied a e: 1. Openness o expe ience: app ecia ion o new emo ions, a s, o new and imagina i e ideas. 2. Conscien iousness: esponsibili y, a endency o sel -discipline, and obedience. 3. Ex a e sion: app ecia ion o a b ead h o ac i i ies. 4. Ag eeableness: sympa hy o iendliness, conce n o social ha mony a he han indi- idualism. 5. Neu o icism: he endency o expe ience nega i e emo ions such as anxie y o dep ession. Following Neuman and Cohen’s 2014 me hodology, we build one ep esen a ion o desc ibing he exis ence o each ac o and ano he ep esen a ion o desc ibing he opposi e o each ac o . In his way, we can analyze ex s w i en by people and, gi en he e idence ound, es ima e he le el o in ensi y o bo h sides o each pe sonali y ac o . Fo example, we migh ack no only he e olu ion o ex a e sion ( he posi i e pole o he 3 d pe sonali y ac o ) bu also he e olu ion o in o e sion, which would be ep esen ed as he nega i e pole o he same pe sonali y ac o . Table 1p esen s he pe sonali y ai s and some wo ds ela ed o each ac o (o o he opposi e o each ac o ). The selec ion o wo ds ep esen ing each pole comes om p e ious s udies on ec o ial seman ics o pe sonali y assessmen , which d ew om a lis o adjec i es ela ed o he Big-5 dimensions (T apnell and Wiggins, 1990). Addi ionally, o he analysis o pe sonali y diso de s, we employ a simila wo d-based app oach o ep esen and ack each diso de . Table 2p esen s he acked pe sonali y dis- o de s and he main adjec i es used o ep esen hem (ex ac ed om Millon e al. (2004)). Fo ou i s analysis, ou objec i e was o s udy how di e en cha ac e is ics o people’s pe sonali ies ha e e ol ed o e he pe iod o he pandemic. To ha end, we exploi he 123 Jou nal o In elligen In o ma ion Sys ems Table 1 Pe sonali y ai s: Ex a e sion (E) Ag eeableness (A) Conscien iousness (C) Neu o icism (N) Openness o expe ience (O) ai s ep esen a i e wo ds E-POSITIVE dominan , asse i e, au ho i a ian, o ce ul, assu ed, con iden , i m, pe sis en E-NEGATIVE ne ous, modes , quie , o celess, a aid, shy, calm, indecisi e A-POSITIVE ende ,gen le,so , kind,a ec iona e,help ul, sympa he ic, iendly A-NEGATIVE c uel, un iendly, nega i e, mean, b u al, inconside a e, insensi i e, cold C-POSITIVE o ganized, o de ly, idy, nea , e icien , pe sis- en , sys ema ic, s aigh , ca e ul, eliable C-NEGATIVE dis ac ed, un eliable, incompe en , wild, ine icien ,disloyal, chao ic, con used,messy, diso ganized N-POSITIVE wo ied, s essed, anxious, ne ous, ea ul, ouchy, guil y, insecu e, es less, emo ional N-NEGATIVE balanced, s able, con iden , ea less, calm, easygoing, elaxed,secu e, com o ed, peace- ul O-POSITIVE philosophical, abs ac , imagina i e, cu ious, e lec i e, li e a y, ques ioning, indi idualis- ic, unique, open O-NEGATIVE na ow-minded,conc e e,o dina y,incu ious, hough less, igno an , uneduca ed, common, con en ional, es ic ed axonomy o pe sonali y ai s p oposed wi hin he Big 5, and we analyze use s’ publica ions acco ding o he i e pe sonali y ac o s (and hei co esponding opposi e poles). The i s s ep o analyzing use da a consis s o con e ing he use ’s wee s o a ec o- ial ep esen a ion. To ob ain his ep esen a ion we employ Sen ence-BERT (Reime s and Gu e ych, 2019) a modi ica ion o he p e- ained BERT ne wo k ha uses ne wo k s uc u es o de i e seman ically meaning ul sen ence embeddings. Sen ence-BERT was ained on he combina ion o he SNLI and Mul i-gen e NLI da ase s. The SNLI collec ion (Bowman e al., 2015) con ains sen ence pai s anno a ed wi h he labels con adic ion, en ailmen , and neu al. The Mul iNLI da ase (Williams e al., 2018) is a collec ion ha con ains sen ence pai s and co e s a ange o gen es o spoken and w i en ex . Sen ence-BERT has been shown o be a Table 2 Pe sonali y diso de s diso de s ep esen a i e wo ds schizoid indi e en , apa he ic, emo e, soli a y dep essi e sad, dep essed, hopeless, gloomy, a alis ic a oidan shy, e lec i e, emba assed, anxious dependen helpless, incapable, passi e, imma u e his ionic d ama ic, seduc i e, shallow, hype ac i e, ain na cissis ic sel ish, a ogan , g andiose, indi e en compulsi e es ained, conscien ious, espec ul, igid pa anoid cau ious, de ensi e, dis us ul, suspicious schizo ypal eccen ic, alien, biza e, absen 123 Jou nal o In elligen In o ma ion Sys ems solid app oach o ep esen sho pieces o ex . We eed each wee o he model o ob ain an embedding ep esen a ion. The ec o ial ep esen a iono eachpe sonali y ac o (o i sopposi epole)isbuil om he wo dlis p esen ed in Table1.Eachwo dis i s passed o helanguagemodel,whichp oduces an embedding ep esen a ion o he wo d. This helps o o mally cap u e he seman ics o he wo d and in e nally ep esen s he po en ial con ex s in which he wo d is used. Nex , he pe sonali y ac o (o i s opposi e) is assigned a ec o ial ep esen a ion ha is he a e age o he ec o s o he wo ds associa ed wi h he ac o . In his way, we can cap u e, o example, aces o neu o icism (N-posi i e ec o ) based on he occu ence o wo ds and exp essions seman ically ela ed o wo ies, s ess, anxie y, ne es, and so o h. To es ima e he deg ee o p esence o a pe sonali y dimension in a wee , we use he cosine simila i y be ween he ex and he dimension’s ep esen a ions. This echnique has p o en o be e y obus in cap u ing seman ic meaning (Laska e al., 2020). Fo ep esen ing he e olu ion o pe sonali y ai s, we plo a e age simila i ies pe week. 3.3 Topic analysis An addi ional le el o analysis o he da a can be o ien ed o ex ac opic-o ien ed signals. To ha end, we ha e pe o med a u he inspec ion o he collec ions wi h Empa h (Fas e al., 2016). Empa h is a ool de eloped a he Uni e si y o S an o d o unde s anding opic signals in la ge collec ions o ex . The o iginal e sion o Empa h was buil om a collec ion o mo e han 1.8 billion wo ds and i es ima es he ela ionship be ween wo ds and ph ases by exploi ing neu al embeddings and sophis ica ed language models. Gi en an inpu ex , Empa h can associa e i o 200 buil -in, p e- alida ed ca ego ies. Fo he ex ac ion o opics, we wo ked wi h Empa h’s de aul ca ego ies and compu ed he le el o p esence o each o hem in he wee s ( o each opic, he lib a y e u ns a sui abili y sco e in [0,1]). Nex , we g ouped he wee s by week and plo ed he a e age alue o he en mos salien ca ego ies. This app oach helps o e eal he mos p ominen opics in he collec ions analyzed. 3.4 Emo ion analysis Fu he e idence abou people’s pe cep ions can be ob ained h ough emo ion analysis, a p ocess ha consis s o iden i ying and analyzing he unde lying emo ions exp essed in ex . Emo ions a e pe asi e among humans and a e s udied in many ields like Psychology and Neu oscience (Canales and Ma ínez-Ba co, 2014). Using emo ions, we can ex ac addi ional clues abou people’s eelings and conce ns. By measu ing emo ions du ing he COVID-19 pandemic, we can gain u he unde s anding on he impac o his di icul pe iod on people’s li es. To pe o m his analysis, we used EmoRoBERTa (Kim and Vossen, 2021), a ans o me language model ained using a da ase o Reddi commen s. The commen s we e labeled ollowing a lis o 28 emo ions. This model has been shown o ou pe o m al e na i e models in mul iple emo ion de ec ion asks and i inco po a es a wide a ie y o emo ions. In Table 3, we p esen some eal examples o wee s classi ied by his model. Again, as o p e ious analysis, we ha e g ouped he wee s by week and we epo he p opo ion o wee s pe week assigned o each emo ion. 123 Jou nal o In elligen In o ma ion Sys ems Table 3 Examples o wee s classi ied by EmoRoBERTa Twee Emo ion “Thank you cm m o his lo ely sha e” g a i ude “Midge Tame Ano he eason I ha e ha guy” ange “de as a ing news om my alma ma e onigh .” sadness “Today is gonna be ha d.” disappoin men “I can’ s op laughing a e alking o hese guys” amusemen 4 Analysis o esul s 4.1 Big-5 Le us indi idually analyze he pa e ns ob ained o each pe sonali y dimension. In his Sec ion, we analyze how pe sonali y ai s a y o e ime. We i s ocus on Ex a e sion, whichconsis so hegene al endency oexpe ienceposi i eemo ions,suchasbeing iendly, li ely, and ac i e. In Fig. 2, we p esen he e olu ion o his ai h ough he pandemic ime. In bo h coun ies (USA and Spain) he nega i e pole o his pe sonali y ac o was dominan (highe line o E-NEGATIVE compa ed o E-POSITIVE). This esul sugges s ha hemes ela ed o ne ousness, ea , o shame (E-NEGATIVE wo ds) we e mo e equen han hose ela ed o E-POSITIVE (dominance, con idence, pe sis ence, e c). This pa e n happens du ing he whole pe iod and, hus, i migh be a gene al ea u e o social media a he han a special ea u e o he COVID-19 pe iod. The USA lines a e a he la and, hus, do no e eal signi ican changes in pa e ns o ex a e sion/in a e sion. The Spain lines (pa icula ly he E-nega i e pa e n) show a mo e agi a ed beha io . Fo example, we can obse e highe e idence o ne es, ea , e c. in he second semes e o 2020. Ano he in e es ing ou come o his analysisis ha , o e all, he p esence o his pe sonali y ac o ( ega dless o he o ien a ion owa ds ex a e sion o in a e sion) seems o be highe in Spain. No e also ha he di e ence be ween ex a e sion (E-POSITIVE) and in a e sion (E-NEGATIVE) is highe in he USA. So, he USA da ase does no appea o con ain oo many ex a e sion/in a e sion- ela ed wee s bu hose wee s ha a e ac ually ela ed o his pe sonali y ac o end o show e idence o ne es, ea , e c. To u he analyze his ac o , we ook he 1000 wee s ha sco ed highes on E- NEGATIVE (highes simila i y o he nega i e pole o his pe sonali y ac o ) and c ea ed a wo d cloud wi h he mos p ominen wo ds (see Fig. 3).6In bo h coun ies, “ne ous” is he dominan wo d, and he plo s also show wo ds ela ed o anxie y and ea . In Spain, we can see ha wo ds like “calm” o “quie ” also had a high deg ee o occu ence. This ma ches wi h he esul s epo ed in Fig. 2, whe e we saw ha , in Spain, he p esence o he E-POSITIVE pole was no insigni ican . Le us mo e o he second pe sonali y ai , Ag eeableness. Ag eeableness is a ai ha desc ibes a pe son’s abili y o pu o he s’ needs be o e hei own. People wi h mo e ag ee- ableness a e mo e likely o be empa he ic and wo k wi h people who need mo e help. Figu e 4 shows he esul s o his pe sonali y ac o . In his g aph, we can see ha , o bo h coun ies, he nega i e pola i y o his pe sonali y ac o is g ea e han he posi i e one. In Spain, we can obse e an inc ease o bo h poles (posi i e and nega i e) in he pe iod o June-July 2020, 6The Spanish wee s ha e been ansla ed in o English and he clouds p esen high- equency wo ds a e ansla ion. 123 Jou nal o In elligen In o ma ion Sys ems Fig. 12 Es ima ed E olu ion o Pe sonali y Diso de s o USA and Spain The second mos p e alen pe sonali y diso de is schizoid. This pe sonali y is a condi ion in which people a oid social ac i i ies and consis en ly shy away om in e ac ion wi h o he s. This ype o beha io is o be expec ed,as he con inemen ules and egula ions issued du ing he pandemic led o isola ion among people. In ac , he blue lines (schizoid pa e n) g ew du ing a co e pa o he pandemic (second semes e o 2020). The hi d dominan pe sonali y diso de is na cissism, a pe sonali y cha ac e ized by a high sense o sel -impo ance. In his case, indi iduals seek a lo o a en ion and a cons an desi e o admi a ion.Du ing hepandemic, hispe sonali ymayha eg own due o he lack o empa hy owa ds o he people. The high p e alence o na cissism also connec s wi h he high es ima e o nega i e ag eeableness desc ibed abo e (high A-NEGATIVE). A-NEGATIVE is indeed connec ed o being unempa he ic and unhelp ul o o he s. In mid-2020 we can see a ma ked inc ease in mos diso de s and, o Spain, we can no ice a sha p decline a he beginning o 2021. USA pa e ns also show a dec ease in mos diso de s in la e 2020/beginning o 2021, bu he decline is smoo he . As a gued abo e, he Spain da ase has ewe da a poin s o he cen al pe iod o 2020 and, hus, he Spain lines a e mo e ola ile. Howe e , he plo o he USA and he plo o Spain essen ially e eal he same ends, ein o cing he insigh s ex ac ed om his analysis. The leas dominan pe sonali y diso de s a e dependen and obsessi e-compulsi e. These a e ela ed o a need o ha e o he people ake ca e o hem o a desi e o ha e e e y hing o ganized. To complemen his analysis, in Table 5, we p esen he wee s wi h he highes alues o he h ee mos p e alen diso de s.7One o he hings we can no ice abou hese wee s is ha , usually, people exp ess hei discom o in sho ex s. Fo example, Na cissis ic diso de is usually associa ed wi h wee s e e ing o someone being sel ish o a ogan . Mo eo e , Appendix Ashows a complemen a y analysis o all ai s in a single plo pe coun y ha allows isualising he ela i e p e alence be ween hem. Table 6gi es an o e all iew o he main ends de i ed om he analysis o pe sonali y diso de s. 7The Spanish wee s ha e been ansla ed in o English and he able p esen s he ansla ed wee . 123 Jou nal o In elligen In o ma ion Sys ems Table 5 Pe sonali y Diso de s Diso de Twee s USA dep ession “Dep essed ” “being dep essed” “oh dea I’m so hopelessly dep essed oday” “Hopeless wo ld” schizoid “apa he ic NY Awhile” “Me mood” “Feeling ha e ul” “yea ning” na cissis ic “Sel ish bas a d” “BE SELFISH” “Sel ish. Absolu ely sel ish.” “Call me sel ish cause I go me and nobody else ” Spain dep ession “Hopeless” “Wha sadness” “Dep essed bu ” “Ch onic sadness.” schizoid “Apa he ic” “D ama mood” “Feeling Alone.” “The indi e ence? I hu o me” na cissis ic “Ayyy Sel ishness” “Sel ish, e y sel ish and bad.” “An a ogan , a ogan , p oud, sel ish...” “Look who’s alking abou indi e en sel ishness...” “Being indi e en nowadays” Example o wee s Table 6 Pe sonali y diso de s analysis The mos p e alen pe sonali y diso de s du ing he ime o he pan- demic we e dep essi e, schizoid, and na cissis ic in bo h coun ies Dep ession is es ima ed o be he mos salien , ollowed by schizoph enia and na cissism The leas dominan pe sonali y diso de s a e dependen and obsessi e-compulsi e Mid-2020: inc ease in mos diso de s in bo h coun ies La e 2020/beginning o 2021: decline in mos diso de s in bo h coun- ies Tendencies (up o down) a e smoo he in USA Main ends ound 123 Jou nal o In elligen In o ma ion Sys ems 4.3 Empa h Nex , we con inue wi h he p esen a ion o he opic-based esul s ob ained wi h Empa h. Figu e 13 epo s, o bo h coun ies, he mos salien opics ex ac ed by Empa h. A i s obse a ion is ha emo ions, ei he posi i e o nega i e, a e he mos p ominen opics. Nega i e emo ions weigh mo e han posi i e emo ions in bo h coun ies. In he USA sample, we can obse e a peak o nega i e emo ions in June 2020 and a highe p esence o nega i e emo ions du ing he second semes e o 2020. The Spain plo shows mo e a iance bu i essen ially e lec s he same end. Compa ing he op 10 Empa h ca ego ies in bo h coun ies we can also obse e o he opics ha a e also common o USA and Spain: iends, op imism, communica ion, pa y, and child en. In he USA plo , mos o he lines a e a he la , and pe haps we can only obse e a sligh dec ease in iends and op imism in he las qua e o 2020. The Spain lines a e mo e chao ic and di icul o g asp. S ill, he e seems o be a highe p esence o con en ela ed o iends, speaking, and op imism du ing he second semes e o 2020. In any case, he mos appa en conclusion om hese plo s is he high p esence o emo ions in bo h coun ies and he occu ence o some o he common opics ( iends, op imism, communica ion, pa y, and child en). We ha e u he analyzed he peaks o nega i e emo ions (see Table 7, which p esen s he high- equency wo ds in he wee s yielding nega i e emo ions). In he USA, we obse e a peak in he i s week o June 2020. This is mainly ela ed o Geo ge Floyd’s dea h, and we obse e he e he p esence o emo ional bea ing wo ds such as “pissed” o “ho ible”. A second nega i e peak, in Augus 2020, is also pa ially ela ed o he Black Li es Ma e mo emen bu we also ind he e nega i e exp essions abou 2020 and he pandemic (“ ucking ha e 2020”, “hell 2020” o “co id”). In Spain, we obse e a sha p ise in nega i e emo ion in he i s week o June 2020. Acco ding o he mos equen wo ds, he heme o Black Li es Ma e is also salien in hese wee s. The second inc ease in nega i e emo ions happened in Oc obe 2020 and, aking in o accoun he mos ecu ing wo ds (e.g., “ eedom”, “c y”, “ i us”, o “go e nmen ”), hese emo ions seem o be ela ed o he co ona i us and o he es ic ions imposed by he go e nmen . Table 8gi es an o e all iew o he main ends de i ed om he analysis pe o med wi h Empa h. Fig. 13 Empa h opics o USA and Spain 123 Jou nal o In elligen In o ma ion Sys ems Table 7 High- equency wo ds du ing he peaks o nega i e emo ions Da e Lis o wo ds USA 1s week o June 2020 sad, ha e, s upid, c azy, ca e, sick, dead, lie, dea h, ho ible, c y, mad, sca ed, wo ied, insane, pain, pissed, shi , guil y, con used, li e, ump, ang y, bi ch, u ah, police, killing, dying, cop, acciden , uck w ong, shocked, makes sick, sho ,s opc ying,s upidpeople,supe , ideo, need s op, s o y,o ice , ucking c azy, ucking sick Las week o Augus 2020 s op, ha e, ha d, ump, sad, w ong, hell, back, lo e, shi , hi , c azy, black, police, li e, wo k, s a e, eason, oday, yea , dead, whi e, igh , s upid, dea h, killed, b eak, lie, alone, ne ada, mad, wo s , nigh , wo ld, incompe ence, egula ions, hu , hope, home, go e nmen , amily, imes, coun y, ame ica, sho , cops, pain, 2020, co id Spain 1s week o June 2020 e dad, ida, mundo, miedo, eeuu, mie da, gobie no, polic, odio, 2020, usa, mue e, culpa, mad e, ju o, casa, iolencia, ascis a, is e, do mi , mo i , e güenza, eliz, locu a, loco, llo ando, democ acia, ump, s op, pais, amilia, causa, bande a, lucha , mue os, acismo, e o is a 2nd week o Oc obe 2020 mal, peo , españa, pu o, mala, mundo, odio, jode , llo a , mo i , is e, mien en, casa, u ina, i us, pa ido, pode , gen uza, libe ad, locos, sola, sin e güenza, ho ible, g i o, gobie no, de echa, clase, payaso, pa idos, nacional, malos, mad id, lucha 4.4 Emo ion analysis Figu e 14 p esen s he dis ibu ion o emo ions. The mos dominan emo ions a e simila in bo h coun ies. Fo example, he emo ions o app o al and admi a ion a e he op 2 emo ions in USA and Spain. O he salien emo ions ha a e common o bo h coun ies a e amuse- men , cu iosi y, lo e, g a i ude, joy, exci emen , sadness, and ange . The ela i e p esence o hese emo ions a ies sligh ly om coun y o coun y. Fo example, he USA sample shows la ge pe cen ages o wee s exp essing amusemen , lo e, sadness, o exci emen , while he Spain sample shows la ge pe cen ages o wee s exp essing g a i ude o ealiza ion. In bo h coun ies, we can obse e ha he leas equen emo ions a e ela ed o emo se, p ide, emba assmen , ne ousness, elie , and g ie . Table 8 Empa h analysis Emo ions, ei he posi i e o nega i e, a e he mos p ominen opics Nega i e emo ions weigh mo e han posi i e emo ions in bo h coun- ies USA, Spain: a peak o nega i e emo ions in June 2020 and a highe p esence o nega i e emo ions du ing he second semes e o 2020 Besides emo ions, o he common opics a e: iends, op imism, com- munica ion, pa y and child en June 2020 nega i e peaks: ela ed o Geo ge Floyd’s dea h O he nega i e peaks e lec ci izens’ conce ns abou co id and es ic ions Main ends ound 123 Jou nal o In elligen In o ma ion Sys ems Fig. 14 Dis ibu ion o emo ions, USA and Spain Le us ocus now on he mos equen emo ions and hei empo al analysis. Figu e 15 p esen s he weekly e olu ion o he op 10 emo ions. The emo ions o admi a ion and app o al we e dominan h oughou he en i e pe iod o he pandemic in he USA. Fo example, in he USA, we obse e wo main peaks o ange , one o hem in June 2020 and ano he one in Janua y 2021. This coincides wi h he p e iously conduc ed opic analysis, whe e we saw an inc ease o nega i e eelings associa ed wi h Geo ge Floyd’s dea h. Ano he emo ion ha shows a signi ican peak is g a i ude, which had a la ge inc ease in bo h coun ies a ound he Ch is mas da es. This sugges s ha people ended o sha e eelings o app ecia ion despi e he di icul ies. In Spain, he emo ions o cu iosi y and sadness domina ed du ing he cen al pa o he pandemic, while he emo ions o g a i ude and admi a ion ended o dec ease conside ably du ing a la ge pa o 2020. The emo ion o exci emen is ela ed o he pe sonali y ac o o “Ex a e sion”. I is in e es ing o no ice ha , o bo h coun ies, we can obse e a d op in Ma ch 2020 and, nex , a sligh endency o inc ease du ing he emainde o he pandemic. This pa e n ma ches wi h ha o “Ex a e sion”, whe e we also saw a d op a he beginning o he pandemic ollowed by a slow eco e y. Table 9gi es an o e all iew o he main ends de i ed om he emo ion analysis. Fig. 15 E olu ion o emo ions, USA and Spain 123 Jou nal o In elligen In o ma ion Sys ems Table 9 Emo ion Analysis The wo mos salien emo ions a e app o al and admi a ion O he equen emo ions in bo h coun ies a e amusemen , cu iosi y, lo e, g a i ude, joy, exci emen , sadness, and ange The leas equen emo ions a e ela ed o emo se, p ide, emba ass- men , ne ousness, elie , and g ie USA: admi a ion & app o al dominan h oughou he en i e pe iod Spain: cu iosi y & sadness high and g a i ude & admi a ion low du - ing he 2nd semes e o 2020 USA: wo main peaks o ange , in June 2020 and Janua y 2021 USA, Spain: d op in exci emen a he beginning o he pandemic ollowed by a slow eco e y Main ends ound 5 Discussion and limi a ions Ou s udy has e ealed in e es ing ends in pe sonali y ac o s, diso de s, emo ions, and he associa ed opics and e en s. The analysis o he wo poles o he 5 pe sonali y dimensions has exposed some consis en indings. Fo example, in bo h samples, we could obse e he p e alence o he nega i e poles o ex a e sion (ne ous, a aid, ...), ag eeableness (c uel, un iendly, ...), and conscien iousness (dis ac ed, un eliable, ...). These h ee poles somehow ep esen he un a o able side o hese 3 Big-5 ac o s. On he o he hand, he p e alence o he nega i e pole o neu o icism (balanced, s able, ...) and he posi i e pole o openness o expe ience (philosophical, abs ac , ...) a e be e news, as hese wo poles eco d a o able aspec s o pe sonali y. Rega ding empo al ends, we ha e obse ed some nega i e changes o e he pe iod analyzed. Fo example, he Spain sample in o ms o an inc ease in neu o icism (N-POSITIVE goes up), in o e sion (E-NEGATIVE goes up), lack o conscien iousness (C- NEGATIVE goes up), and closeness o expe ience (O-NEGATIVE goes up) in he second semes e o 2020. The USA ends also e eal some conce ning e olu ion du ing 2020, wi h an inc ease in neu o icism (N-POSITIVE goes up) and lack o conscien iousness (C- NEGATIVE goes up), and a dec ease in ag eeableness (A-NEGATIVE goes up). Ou analysis also sugges s ha he mos p e alen pe sonali y diso de s a e dep ession, schizoph enia, and na cissism (in bo h coun ies). The p esence o signs o hese diso de s inc eased in mid-2020 and had a decline in la e 2020/beginning o 2021. The analysis o Empa h opics showed a high p esence o emo ional publica ions, wi h a gene al p e alence o nega i e emo ions o e posi i e ones, and he occu ence o peaks o nega i e emo ions ela ed o COVID-19 and Geo ge Floyd’s dea h. By p ocessing he da a agains a ine-g ained ca ego iza ion o 28 ypes o emo ions, we obse ed ha admi a ion and app o al a e he mos dominan emo ions. S ill, we also see he e some signs o conce n. Fo example, in he USA, he e we e wo peaks o ange (June 2020 and Janua y 2021) and a dec ease in exci emen a he beginning o he pandemic. In Spain, we also obse e he same dec ease in exci emen and, u he mo e, highe signs o sadness and lowe signs o g a i ude and admi a ion du ing he second semes e o 2020. The s udy p esen ed he e has po en ial limi a ions. The es ima es epo ed a e based on wo samples o wee s ex ac ed a speci ic geog aphical loca ions. The wo collec ions a e la ge bu , s ill, a e only e lec i e o he p eoccupa ions o speci ic segmen s o he USA o Spain popula ions. Fu he mo e, we canno claim ha his gi es a pic u e o he en i e popula ion in hese a eas, because he samples a e ob iously biased owa d people who a e 123 Jou nal o In elligen In o ma ion Sys ems ac i e on a speci ic social media pla o m. In any case, we i mly belie e ha his ype o social media analysis p o ides aluable insigh s in o a signi ican pa o he popula ion. The Spain sample, because o echnical di icul ies wi h he ex ac ion ool, has ewe da a poin s du ing he second pa o 2020. S ill, we ha e ocused ou analysis on he common ends and we a e awa e ha some s ong luc ua ions obse ed in he plo s a e ela ed o he size o some weekly samples. Ano he main limi a ion comes om he es ima ion me hods. In hese collec ions, we do no ha e a g ound u h o pe sonali y labels, diso de s, o emo ions. We a he ely on epu ed ools, such as deep linguis ic models, emo ion de ec o s, Empa h, and so o h. This undoub edly in oduces noisy es ima ions abou he p esence o signs o pe sonali y, diso de s, opics, o emo ions. S ill, hese ools ha e been es ed elsewhe e and his ype o socialmedia moni o ing migh e o speci icda apoin sbu ends o be eliable a iden i ying ela i e ends and hei empo al e olu ion. 6 Conclusions In his pape , we ha e c ea ed and made a ailable o he communi y wo la ge collec ions o wee s published du ing he agi a ed pe iod o he pandemic. These se s o wee s come om wo coun ies: USA and Spain. Gi en his la ge sample o publica ions, we ha e pe o med a compa a i e analysis o pe sonali y ac o s and diso de s, opics, and emo ions. We ha e been able o obse e how di e en e en s ha e in luenced he esponse o he people and we ha e iden i ied speci ic ime ames when emo ions o pe sonali y ac o s su e ed subs an ial changes. Fo example, nega i i y a ose no only because o COVID-19 bu also because o o he majo e en s ha happened in hese wo u bulen yea s. We can conside social ne wo ks as a well-es ablished sou ce o in o ma ion om which we can ex ac aluable knowledge and d aw a pa ial bu use ul iew o he eelings o he popula ion. The syne gies be ween heal h and web/social mining ha e been explo ed o a numbe o yea s (e.g., o designing new sc eening echnologies). Howe e , we s ill need o iden i y he speci ic ole ha echnology can play in suppo ing men al heal h p o essionals. This ype o applica ion is s ill in i s in ancy and ou pape has add essed some moni o ing ac i i ies ha could be pa o u u e pe sonali y/sen imen /emo ion analyze s. Fo example, a gi en public ins i u ion migh be in e es ed in acking ce ain segmen s o he popula ion and ale abou he onse o some ypes o isks. In he nea u u e, we plan o con ac wi h ele an s akeholde s, such as ained psychologis s and o he men al heal h p o essionals, o alida e his echnology and o explo e u u e ways o exploi a ion. As u u e wo k, we a e also in e es ed in expanding he s udy o o he da a sou ces and pla o ms. I would be s imula ing o compa e mul iple in o ma ion sou ces and social ne - wo ks o ex ac he in ensi y o emo ions, pe sonali y aspec s, and opics and see how hey di e o e sou ces. Appendix A: Ex a pe sonali y ai analysis To complemen he pe sonali y ai analysis and unde s and which ac o is mo e p e alen , Figs.16and17p esen he e olu iono all ai s h ough ime.Wecansee ha henega i epole is dominan in bo h coun ies and ha he ai s ela ed o conscien iousness and ex a e sion ha e he g ea es p esence. 123 Jou nal o In elligen In o ma ion Sys ems Fig. 16 Big-5 analysis du ing he COVID-19 pandemic o USA Fig. 17 Big-5 analysis du ing he COVID-19 pandemic o Spain 123 Jou nal o In elligen In o ma ion Sys ems Au ho Con ibu ions Concep ualiza ion: [all au ho s]; Me hodology: [all au ho s]; Fo mal analysis: [all au ho s]; In es iga ion: [all au ho s]; Da a cu a ion: [Julian Sabo ido-Pa i˜ ño, Ma cos Fe nández-Pichel, Ma io Ez a A agón]; Valida ion: [Ma cos Fe nández-Pichel, Ma io Ez a A agón]; W i ing - o iginal d a p epa a- ion: [Ma io Ez a A agón]; W i ing - e iew and edi ing: [Da id E. Losada, Ma cos Fe nández-Pichel, Ma io Ez a A agón]; Supe ision: [Da id E. Losada, Ma io Ez a A agón]; P ojec adminis a ion: [Da id E. Losada]. Funding Open Access unding p o ided hanks o he CRUE-CSIC ag eemen wi h Sp inge Na u e. The au ho s hank he suppo ob ained om: i) p ojec PLEC2021-007662 (MCIN/AEI/10.13039/50 1100011033, Minis e io de Ciencia e Inno ación, Agencia Es a al de In es igación, Plan de Recupe ación, T ans o mación y Resiliencia, Unión Eu opea-Nex Gene a ionEU), ii) p ojec PID2022-137061OB-C22 (Minis e io de Ciencia e Inno ación, Agencia Es a al de In es igación, P oyec os de Gene ación de Conocimien o; supppo ed by he Eu opean Regional De elopmen Fund) and iii) Conselle ía de Educación, Uni e sidade e Fo mación P o esional (acc edi a ion 2019-2022 ED431G-2019/04, ED431C 2022/19) and he Eu opean Regional De elopmen Fund, which acknowledges he CiTIUS-Resea ch Cen e in In elligen Technologies o he Uni e si y o San iago de Compos ela as a Resea ch Cen e o he Galician Uni e si y Sys em. Da a and Code a ailabili y The da ase s and codes gene a ed du ing and/o analyzed du ing he cu en s udy a e a ailable om he co esponding au ho on easonable eques . Decla a ions E hics app o al No applicable Consen o pa icipa e No applicable Consen o publica ion No applicable Compe ing in e es s The au ho s ha e no compe ing in e es s o decla e ha a e ele an o he con en o his a icle. Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a c edi line o he ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommons.o g/licenses/by/4.0/. Re e ences Ahmad, A. R., & Mu ad, H. R. (2020). The impac o social media on panic du ing he co id-19 pandemic in i aqi ku dis an: Online ques ionnai e s udy. J Med In e ne Res, 22(5), e19556. h ps://doi.o g/10.2196/ 19556 Ainley, E., Wi wicki, C., Talle , A., e al. (2021). Using wi e commen s o unde s and people’s expe iences o uk heal h ca e du ing he co id-19 pandemic: Thema ic and sen imen analysis. J Med In e ne Res. h ps://doi.o g/10.2196/31101 Alhuzali, H., Zhang, T., & Ananiadou, S. (2022). Emo ions and opics exp essed on wi e du ing he co id-19 pandemic in he uni ed kingdom: Compa a i e geoloca ion and ex mining analysis. J Med In e ne Res, 24(10). h ps://doi.o g/10.2196/40323 Ba ick, M. R., & Moun , M. K. (1991). The big i e pe sonali y dimensions and job pe o mance: A me a- analysis. Pe sonnel Psychology, 44, 1–26. h ps://doi.o g/10.1111/j.1744-6570.1991. b00688.x Bo ah, A. (2023). De ec ing co id-19 accine hesi ancy in india: a mul imodal ans o me based app oach. Jou nal o In elligen In o ma ion Sys ems, 60(1), 157–173. h ps://doi.o g/10.1007/s10844-022-00745- 1 123 Jou nal o In elligen In o ma ion Sys ems Bowman, S. R., Angeli, G., Po s, C., e al. (2015). A la ge anno a ed co pus o lea ning na u al language in e ence.In: P oceedings o he2015 Con e enceon Empi icalMe hodsin Na u al LanguageP ocessing. Associa ion o Compu a ional Linguis ics. Lisbon, Po ugal. h ps://doi.o g/10.18653/ 1/D15-1075 Canales, L., & Ma ínez-Ba co, P. (2014). Emo ion de ec ion om ex : A su ey. P ocessing in he 5 h In o ma ion Sys ems Resea ch Wo king Days (JISIC).h ps://doi.o g/10.3115/ 1/W14-6905 Di, X., Li a, W., Zheng, H., e al. (2018). Deep lea ning-based pe sonali y ecogni ion om ex pos s o online social ne wo ks. Applied In elligence, 48,. h ps://doi.o g/10.1007/s10489-018-1212-4 Fas , E., Chen, B., & Be ns ein, M. S. (2016). Empa h: Unde s anding opic signals in la ge-scale ex . In: P oceedings o he 2016 CHI Con e ence on Human Fac o s in Compu ing Sys ems. Associa ion o Compu ing Machine y. New Yo k, NY, USA, CHI ’16. h ps://doi.o g/10.1145/2858036.2858535 Funde , D.C. (1997). The pe sonali y puzzle. W W No on & Co Goldbe g, L. R. (1993). The s uc u e o pheno ypic pe sonali y ai s. Ame ican Psychologis .h ps://doi.o g/ 10.1037/0003-066X.48.1.26 Gup a, R. K., Vishwana h, A., & Yang, Y. (2020). COVID-19 wi e da ase wi h la en opics, sen imen s and emo ions a ibu es. CoRR abs/2007.06954. a Xi :2007.06954 John, O. P., & S i as a a, S. (1997). The big i e ai axonomy: His o y, measu emen , and heo e ical pe spec i es. L. A. Pe in & O. P. John (Eds.), Handbook o pe sonali y: Theo y and esea ch Kim, T., & Vossen, P. (2021). Emobe a: Speake -awa e emo ion ecogni ion in con e sa ion wi h obe a. h ps://doi.o g/10.48550/ARXIV.2108.12009 Laska , M. T. R., Huang, X., & Hoque, E. (2020). Con ex ualized embeddings based ans o me encode o sen ence simila i y modeling in answe selec ion ask. In: P oceedings o The 12 h Language Resou ces and E alua ion Con e ence Leona di, S., Mon i, D., Rizzo, G., e al. (2020) Mul ilingual ans o me -based pe sonali y ai s es ima ion. In o ma ion 11(4). h ps://doi.o g/10.3390/in o11040179 Lyu, H., Chen, L., Wang, Y., e al. (2020). Sense and sensibili y: Cha ac e izing social media use s ega ding he use o con o e sial e ms o co id-19. IEEE T ans Big Da a, 7, 952–960. h ps://doi.o g/10.1109/ TBDATA.2020.2996401 Millon, T., Millon, C., & Meaghe , S. (2004). B ie desc ip ion o he ou een pe sonali y diso de s o dsm-iii. DSM-III-R: Tech. ep. Neuman, Y., & Cohen, Y. (2014). A ec o ial seman ics app oach o pe sonali y assessmen . Scien i ic epo s, 4(1), 1–6. h ps://doi.o g/10.1038/s ep04761 O ganiza ion WH (2020). Impac o co id-19 on people’s li elihoods, hei heal h, and ou ood sys- em. h ps://www.who.in /news/i em/13-10-2020-impac -o -co id-19-on-people’s-li elihoods- hei - heal h-and-ou - ood-sys ems. Accessed 06 Jan 2023 O ganiza ion WH (2023). WHO co ona i us disease (COVID-19) dashboa d. h ps://co id19.who.in /. Accessed 06 Jan 2023 Reime s, N., & Gu e ych, I. (2019). Sen ence-be : Sen ence embeddings using siamese be -ne wo ks. In: P oceedings o he 2019 Con e ence on Empi ical Me hods in Na u al Language P ocessing. Associa ion o Compu a ional Linguis ics. a Xi :1908.10084 Reu e s (2022). Digi al news epo 2022. h ps:// eu e sins i u e.poli ics.ox.ac.uk/digi al-news- epo /2022. Accessed 06 Jan 2023 de-la Rosa, G. R., Jiménez-Salaza , H., Villa o o-Tello, E., e al. (2023). A lexical-a ailabili y-based amewo k om sho communica ions o au oma ic pe sonali y iden i ica ion. Cogni i e Sys ems Resea ch, 79, 126–137. h ps://doi.o g/10.1016/j.cogsys.2023.01.006 Se , E., Okan, O., Özbilen, A., e al. (2022). Linking co id-19 pe cep ion wi h socioeconomic condi ions using wi e da a. IEEE T ansac ions on Compu a ional Social Sys ems, 9(2), 394–405. h ps://doi.o g/ 10.1109/TCSS.2021.3089657 Si, M. Y., Su, X. Y., Jiang, Y., e al. (2021). P e alence and p edic o s o p sd du ing he ini ial s age o co id-19 epidemic among emale college s uden s in china. INQUIRY: The Jou nal o Heal h Ca e O ganiza ion, P o ision, and Financing 58, 00469580211059953. h ps://doi.o g/10.1177/00469580211059953 T apnell, P. D., & Wiggins, J. S. (1990). Ex ension o he in e pe sonal adjec i e scales o include he big i e dimensions o pe sonali y. J Pe s Soc Psychol, 59,. h ps://doi.o g/10.1037/0022-3514.59.4.781 Umai , A., & Mascia i, E. (2023). Sen imen al and spa ial analysis o co id-19 accines wee s. Jou nal o In elligen In o ma ion Sys ems, 60(1), 1–21. h ps://doi.o g/10.1007/s10844-022-00699-4 Voh a, A., & Ga g, R. (2023). Deep lea ning based sen imen analysis o public pe cep ion o wo king om home h ough wee s. Jou nal o In elligen In o ma ion Sys ems, 60(1), 255–274. h ps://doi.o g/10. 1007/s10844-022-00736-2 Williams, A., Nangia, N., & Bowman, S. (2018). A b oad-co e age challenge co pus o sen ence unde s and- ing h ough in e ence. In: P oceedings o he 2018 Con e ence o he No h Ame ican Chap e o he Associa ion o Compu a ional Linguis ics: Human Language Technologies, Volume 1 (Long Pape s). 123