This a icle is dis ibu ed unde he e ms o he
C ea i e Commons A ibu ion 4.0 In e nacional License
ins i u ua
enp esa
Ins i u o de Economía Aplicada a la Emp esa
Managemen Le e s / Cuade nos de Ges ión
jou nal homepage: h ps://ojs.ehu.eus/index.php/CG
ISSN: 1131-6837 / e-ISSN: 1988-2157
Managemen Le e s
Cuade nos de Ges ión
Enp esa Ins i u ua, UPV/EHU
Conocimien o en Ges ión/Managemen Knowledge
Volume 25 / Numbe 1 (2025) • ISSN: 1131-6837 / e-ISSN: 1988-2157
h p://www.ehu.eus/cuade nosdeges ion/ e is a/es/
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es
h ough Deep Lea ning o online e iews
Mejo a de la sa is acción del clien e median e la iden i icación de allos en los mic oelemen os del
se icio ho ele o median e el Deep Lea ning de las eseñas online
Se gey Kazako *, Ped o Cues a-Valiño
a
, Ve a Bu kouskaya
b
, Daniel Mu a sky
c
a Uni e si y o Alcalá. Facul y o Economic, Business and Tou ism Sciences. Plaza de la Vic o ia, 2 28802 Alcalá de Hena es (Mad id), Spain–ped o.cues [email p o ec ed]–
h ps://o cid.o g/0000-0001-9521-333X
b G adua e School o Business, HSE Uni e si y. Tecnocampus, Uni e si ad Pompeu Fab a. Ca e d'E nes Lluch, 32, 08302 Ma a ó, Ba celona, Spain– bu kouskaya@
ecnocampus.ca – h ps://o cid.o g/0000-0002-6963-3872
c Uni e si y o Wes Sco land, School o Business and C ea i e Indus ies. Impo Building, 2 Clo e C es, London E14 2BE, Uni ed Kingdom–P o .mu a sk[email p o ec ed]m– h ps://
o cid.o g/0000-0001-8401-1476
* Co esponding au ho : Uni e si y o Alcalá. Facul y o Economic, Business and Tou ism Sciences. Plaza de la Vic o ia, 2 28802 Alcalá de Hena es (Mad id),
Spain–se gey.kazako @uah.es–h ps://o cid.o g/0000-0001-5532-5791
ARTICLE INFO
Recei ed 21 Feb ua y 2024,
Accep ed 18 Aug 2024
A ailable online 21 Ma ch 2025
DOI: 10.5295/cdg.242191sk
JEL: M31
ABSTRACT
This s udy ho oughly examines o en-o e looked mic o-se ice elemen s wi hin he b oade spec um o ho el
se ices, aiming o imp o e hospi ali y se ices and ensu e gues sa is ac ion. To achie e his, his esea ch de el-
oped a me hodological amewo k, in eg a ing (a) he VADER ex sen imen analysis amewo k, (b) a obus
logis ic eg ession p ocedu e o pinpoin speci ic ho el se ice componen s culp i o gues us a ion, and (c)
he applica ion o seman ic ne wo k analysis o yield gues insigh s con ex ualised wi hin he ealm o unde pe -
o ming ho el se ice mic o-elemen s.
Resea ch indings highligh i y speci ic se ice mic o-elemen s iden i ied as igge s o nega i e sen imen and
subsequen deg ees o diminished gues sa is ac ion.Fu he mo e, his s udy zooms in o he op en unde pe -
o ming se ice mic o-elemen s by employing seman ic ne wo k analysis o unco e he oo s o ypical gues
us a ions wi h hei ho el expe iences. Though iden i ied wi hin ho el e iews, ce ain se ice mal unc ions
ha e ele ance wi hin he b oade domain o des ina ion managemen .
The ou comes o his s udy sugges a aluable esou ce o manage s in de ec ing and ec i ying inadequa ely
pe o ming ho el se ice mic o-elemen s, which a e pi o al o ele a ing gues sa is ac ion wi hin hei espec i e
ho el p ope ies. Addi ionally, he indings p o ide impe us o ho el and des ina ion manage s o implemen
ailo ed s a egies o inc ease gues sa is ac ion ac oss ho els and des ina ions.
Keywo ds: ho el se ice elemen s, online e iews, na u al language p ocessing, big da a, ou is sa is ac ion policy,
eWOM.
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
72 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
RESUMEN
Es e es udio examina en p o undidad los elemen os de mic ose icios a menudo pasados po al o den o del am-
plio espec o de se icios ho ele os, con el obje i o de mejo a la hospi alidad y ga an iza una mayo sa is acción
de los huéspedes. Pa a log a lo, se desa olló un ma co me odológico que in eg a (a) el análisis de sen imien o de
ex o VADER, (b) un p ocedimien o obus o de eg esión logís ica pa a iden i ica los componen es especí icos
del se icio ho ele o que causan us ación a los huéspedes, y (c) el análisis de edes semán icas pa a gene a
in o mación ma izada sob e los huéspedes, con ex ualizada den o del ámbi o de los mic oelemen os de se icio
ho ele o de bajo endimien o.
Los esul ados de la in es igación des acan cincuen a mic oelemen os de se icio especí icos que desencadenan
sen imien os nega i os y una disminución subsecuen e en la sa is acción de los huéspedes. Además, es e es udio
se en oca en los diez mic oelemen os de se icio de meno endimien o, u ilizando el análisis de edes semán icas
pa a descub i las causas p incipales de las us aciones comunes de los huéspedes con sus expe iencias ho ele as.
Algunos allos en el se icio, aunque se iden i ican en las eseñas de ho eles, son ele an es ambién en el ámbi o
más amplio de la ges ión de des inos.
Los hallazgos de es e es udio sugie en un ecu so alioso pa a los ge en es en la de ección y co ección de mi-
c oelemen os de se icio ho ele o que uncionan de mane a inadecuada, undamen ales pa a ele a la sa is acción
de los huéspedes en sus espec i as p opiedades ho ele as. Además, los esul ados incen i an a los ge en es de
ho eles y des inos a implemen a es a egias pe sonalizadas des inadas a mejo a la sa is acción de los huéspedes
en odos los ho eles y des inos.
Palab as cla e: elemen os de se icio ho ele o, eseñas online, p ocesamien o del lenguaje na u al, big da a, polí i-
ca de sa is acción del u is a, eWOM.
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es h ough Deep Lea ning o online e iews 73
1. INTRODUCTION
Online e iews ha e become a i al channel o ho el gues s
o sha e hei a el expe iences (Xie e al., 2014; Casalo e al.,
2015). These e iews a e c ucial in he cus ome jou ney, o en
ep esen ing he inal s age in ou ism and hospi ali y se ings
(Cha e jee, 2020). They p o ide aluable con en o p ospec-
i e ho el cus ome s (Repo ienė & Pažė ai ė, 2023) and ep e-
sen elec onic wo d-o -mou h (eWOM), a digi al ex ension o
he adi ional wo d-o -mou h (WOM) concep in he domain
o con en ional ma ke ing (Bu kouskaya e al., 2020; Le e al.,
2023). As such, online e iews signi ican ly in luence ho el
booking decisions, p esen ing immense oppo uni ies o hos-
pi ali y esea che s o explo e a ious ace s o gues beha iou
(Yang e al., 2020). Consequen ly, ad ancemen s in ICT, easy
access o online e iews as esea ch da a, and machine lea n-
ing echnologies ha e ueled a p oli e a ing s eam o big da-
a-based esea ch in ou ism and hospi ali y (Cues a-Valiño
e al., 2020).
In hei e iews, ho el gues s equen ly commen on he se -
ices hey encoun e , including ho el ooms and acili ies, pe -
sonnel, se ices, loca ion, and ood (Nie e al., 2020; Za ezadeh
e al., 2022). Simul aneously, ho el gues s commonly e alua e
hese se ice mac o- ac o s and ocus on speci ic se ice mi-
c o-elemen s wi hin he abo e-no ed ca ego ies. Fo ins ance,
when assessing a ho el oom, gues s o en highligh aspec s like
cleanliness, bed quali y, Wi-Fi, and ba h oom ameni ies (Luo
e al., 2021).
Gi en hei impac on o e all gues e alua ions, ho el se ice
mic o-elemen s a e meaning ul and hus equi e mo e conside -
a ion in hospi ali y ma ke ing li e a u e. We de ine ho el se ice
mic o-elemen s as speci ic and de ailed ace s o he hospi ali-
y se icescape ha collec i ely shape he o e all gues expe i-
ence. Al hough p e ious esea ch has examined mic o-elemen s’
e ec s on gues sa is ac ion (Hu e al., 2020), i o en used he
e m’ a ibu es’ wi hou dis inguishing be ween di e en le els
o se ice elemen s.
In mo e de ail, ecen s udies ha e ope a ed wi h se ice mac-
o ac o s, e e ed o as ‘a ibu es’, o all elemen s o he ho el’s
se icescape wi hou making a dis inc ion. Fo ins ance, p io li -
e a u e has posi ioned such se ice elemen s as ‘bed’, ‘ba h oom’,
‘in e ne ’ and ‘ oom’ o ‘d ink’, ‘lobby’ and ‘ba ’ on he same le el.
Howe e , he e is an appa en hie a chy whe e ‘ oom’ is a highe
o de elemen comp ising ‘bed’, ‘ba h oom’, and ‘in e ne ’. A he
same ime, ‘d ink’ belongs o ‘ba ’, which sequen ially i s in he
‘lobby’, a la ge -scale se ice a ibu e. Wi hou making a mul-
i-le el axonomy be ween he sensual dimensions o he ho el
se ice elemen s, i becomes challenging o ho el manage s o
de e mine an accu a e, some imes hidden eason o gues s’ dis-
appoin men wi h he hospi ali y se ice. Mo eo e , as he pe -
cep ion o he highe -o de se ice elemen s ( oom, o ins ance)
deno es a complex combina ion and likely implies a eg essed
sum o he lowe -le el elemen s (bed, ai condi ioning, u ni u e,
e c.), conside ing gues s’ ho el e alua ion should be mul i ace ed
and non-linea gi en om his pe spec i e also.
Despi e some amoun o ex an esea ch on gues sa is ac-
ion using ad anced echniques o ex mining and na u al lan-
guage p ocessing (NLP) applied o online gues e iews as a da a
sou ce(e.g., Hu & Yang, 2021; Shin e al., 2021), he abo e-no ed
limi a ions gene a e a signi ican esea ch gap ha is awai ing ac-
ademic a en ion. Apa om con usion due o he no ed se ice
elemen s ailu es examina ion wi hou conside ing he elemen s’
hie a chy, p io s udies ha e also o en been limi ed o speci ic
des ina ions, such as China. Consequen ly, he li e a u e mus
s ill add ess he esea ch indings' gene alisa ion issues. Mo eo-
e , s udies in he ield o hospi ali y cus ome beha iou should
also p o ide be e in-dep h and mo e conc e e insigh s in o he
easons o nega i e sen imen s disco e ed in gues e iews.
Add essing nega i e cus ome eedback is c ucial, as neg-
a i ely in ed e iews signi ican ly impac ho el pe cep ions o
p ospec i e cus ome s (Hu e al., 2020). G ounded in he accu-
mula ed esea ch, i s abo e-no ed accomplishmen s and gaps
de e mined in he ex an body o li e a u e, his s udy poses he
ollowing esea ch ques ion:
RQ: Acco ding o online e iews, which speci ic ho el se ice
mic o-elemen s gene a e he mos nega i e sen imen ,
lead o poo cus ome expe iences, and dis up gues sa -
is ac ion?
To add ess he posed RQ, his explo a o y s udy aims o (1)
use deep lea ning and ex mining o iden i y ho el se ice mi-
c o-elemen s ha con ibu e o nega i e gues sen imen in on-
line e iews ac oss ele en popula ou is des ina ions and (2)
de e mine he ailing se ice mic o-elemen s and explo e he
con ex o hese poo ly deli e ed se ices o unde s and he un-
de lying causes o gues dissa is ac ion. By add essing he pos-
i ed RQ, his s udy makes wo signi ican con ibu ions: Fi s ,
in he ealm o hospi ali y ma ke ing li e a u e, i de e mines
he op ailing ho el se ice mic o-elemen s and hei con ex s
o e looked by p io s udies. Second, his esea ch p o ides ho el
manage s wi h in o ma ion on speci ic se ice laws equi ing
hei a en ion. Also, his s udy bes ows de ailed p ac ical ec-
ommenda ions o imp o e se ice quali y and gues expe iences.
The emaining manusc ip is o ganised as ollows. Fi s , we
e iew he li e a u e ha unde pins ou esea ch a ionale and
suppo s he posed RQ. Nex , his pape p esen s he esea ch
me hodology, explica ing he de eloped app oach o da a col-
lec ion and analysis p ocedu es. Then, he na a i e p esen s he
indings and discusses he implica ions o hospi ali y ma ke ing
heo y and ho el managemen p ac ices. Finally, we ou line e-
sea ch limi a ions and p opose u u e esea ch di ec ions, con-
cluding wi h a summa y o ou key indings.
2. LITERATURE REVIEW
2.1. Online gues e iews and ho el se ice ailu es
As p e iously no ed, se e al esea che s indica ed ha
online gues e iews a e a p ac ical da a sou ce o collec ing
gues s' insigh s o e alua e a cus ome 's pe cep ion o he ho el
se ice quali y (Be ezina e al., 2016; Song e al., 2022; Za eza-
deh e al., 2022). Ho el gues s sha e hei expe iences wi h o h-
e use s by pos ing e iews o social media, po als o online
a el agencies (OTAs) o e iew-specialised pla o ms (Khan
e al., 2022; Viei a e al., 2023). Online e iews deno e a piece
o use -gene a ed con en (UGC) commonly comp ising bo h
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
74 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
quali a i e ( e iew ex ) and quan i a i e (gues -assigned ho el
a ing sco e) da a (Ma iani e al., 2019). As a aluable sou ce
o public da a, online e iews help enac analy ical echniques
such as social media lis ening, big da a analy ics and ex min-
ing (Hu & Yang, 2021).
Nega i e e iews pos ed by ho el gues s a e o pa icula
esea ch in e es because hey allow ho el manage s o iden i y
se ice ailu es and, hus, ind oppo uni ies o imp o e se ices
deli e ed by hei p ope y (Son e al., 2022). Resea che s a gue
ha ho el se ice ailu e e e s o si ua ions when he ho el se -
ices do no comply wi h he gues s' expec a ions, leading o a
signi ican de ia ion om he expec ed se ice s anda ds (Sann
e al., 2021). An ex an body o amassed esea ch has used ex
mining and de e mined se e al ac o s gues s associa e wi h ho-
el se ice ailu es (e.g., Huang e al., 2022; Nie e al., 2020; Ying
e al., 2020).
In his ein, esea che s i s poin ed o oom cleanliness as
he c i ical mic o-elemen p edisposing nega i e low gues sa is-
ac ion (Pa k e al., 2019). P io esea ch has also no ed ho el a-
cili ies in online e iews conce ning se ice ailu es (Ying e al.,
2020). Fu he mo e, acco ding o he li e a u e, gues s indica e
cus ome se ice p o ided by hospi ali y pe sonnel in hei neg-
a i e e iews as a ac o dis up ing hei sa is ac ion wi h he ho-
el (Nie e al., 2020).
2.2. Ho el se ice mic o-elemen s
Va ious ma ke ing esea che s ha e explo ed how cus ome s
pe cei e se ice elemen s ac oss ecei ing cus ome expe iences
and sa is ac ion wi h se ice p oduc s (Bueno e al., 2019; Roy,
2018). In hospi ali y li e a u e, amassed esea ch has de e mined
d i e s in luencing gues sa is ac ion wi h ho el se ices (Lee
e al., 2020). In his ein, as p e iously no ed, se e al esea ch-
e s poin ed o he ho el oom as he c i ical ac o p edisposing
nega i e sen imen and, hus, low gues sa is ac ion (Padma &
Ahn, 2020; Pa k e al., 2019). Nex , La inopoulos (2020) a ib-
u ed ho el loca ion and ex e io as a highly in luen ial elemen
o he ho el se ice. Fu he mo e, gues s indica e ood quali y as
a ac o sculp ing hei sa is ac ion wi h he ho el (Philips e al.,
2017; Za ezadeh e al., 2022).
Simul aneously wi h he es ablished esea ch ocusing on
he o e a ching se ice ac o s ele an o gues sa is ac ion, a
g owing body o li e a u e del es in o he ine o mic o-ele-
men s o ho el se ice elemen s. In his ega d, Nie e al. (2020)
e ealed sleep quali y as a ho el oom’s mic o-elemen , which
is signi ican o he gues s. La e , Hu e al. (2021) echoed hese
indings bu un eiled an ex ended ange o ho el mic o-se ices
in luencing gues sa is ac ion, including oom a e, lobby ba , e-
cep ion s a , b eak as , and wi- i signal quali y. Simul aneously,
Luo e al. (2021) no ed se ice mic o-elemen s o he wi- i sig-
nal, ai condi ioning, bed, noise, owels, hai d ye s, and slippe s
p io ly examined as bigge ‘ oom’ o ‘ oom acili ies’ ac o by
o he esea che s (Alnawas & Hemsley-B own, 2018). Mo e e-
cen ly, Song e al. (2022) used he La en Di ichle Alloca ion
(LDA) algo i hm o ex ac opics om mo e han 50 000 gues
e iews. In ha s udy, LDA ex ac ed i e opics ha ma e o
gues sa is ac ion, including Se ice, Room, Cleanliness, Loca-
ion, and Value (Song e al., 2022). P io esea che s ha e widely
implemen ed sen imen analysis, discussed below, as a ex min-
ing echnique o ob ain hese indings. By employing sen imen
analysis, hey ex ac ed nega i e gues e iews i s and, by ap-
plying addi ional sc u iny, ound he abo e-no ed ho el se ice
ailu e ac o s ha commonly educe gues sa is ac ion wi h he
ho el (Huang e al., 2022).
2.3. Sen imen analysis in hospi ali y esea ch
Sen imen analysis is a ex mining p ocedu e esea che s use
o na ow down a ich pale e o human emo ions, including hap-
piness, pleasu e, apa hy, annoyance, age, e c., in o h ee dis inc
ca ego ies o nega i e, neu al o posi i e pola i y (Ki ilenko e al.,
2018; Luo e al., 2021). Online gues e iews deno e a big da a
solu ion o hospi ali y esea che s ha allows hem o gauge
e iew sen imen mo e accu a ely and depic e iew pola i y
wi h a p ecise nume ic alue. Online gues e iews benchma k
a subs an ial mo e o wa d in hospi ali y analy ics and g adu-
ally eplace con en ional me hods o da a collec ion, e.g. su -
eys, equi ed o ob aining cus ome sen imen da a (Hu e al.,
2020). By applying he ex mining echnique and undamen al
sen imen analysis p ocedu es, hospi ali y esea che s look o
speci ic wo ds o comple e ph ases in cus ome e iews which
de e mine a gues ’s posi i e o nega i e sen imen . Acco ding-
ly, se e al s udies ha e exhibi ed isually imp essi e wo d cloud
models g ounded in con en equency analysis o e iews (Hu &
Yang, 2021; Shin e al., 2021).
Sen imen analysis is a pa o he Na u al Language P o-
cessing (NLP) algo i hms amily, which is a ami ica ion o he
b oade machine lea ning p ocedu es cons ella ion. Thanks
o i s e sa ile bene i s, lexicon-based sen imen analysis has
ecei ed ecogni ion as a obus me hodology in academia. I
has se he momen um o an eme ging ou ism and hospi-
ali y esea ch s eam. In his ein, Yada and Roychoudhu y
(2019) applied sen imen analysis o in es iga e ho el a ib-
u es ha a elle s pe cei e as essen ial when planning hei
ip in di e en a el modes (leisu e, business, single, couple,
amily, e c.). Simila ly o hese indings, a s udy by Be ezina
e al. (2016) e ealed common ca ego ies used in posi i e and
nega i e e iews by applying ex mining echniques o ana-
lyse online cus ome e iews. Mo eo e , se e al s udies u ilised
sen imen analysis o de elop p edic i e models o compu ing
nume ical a ing sco es missing in he e iew da ase (Gee ha
e al., 2017; Kim & Im, 2018).
Nex , sen imen analysis demons a ed i s cogen capabili ies
o de e mine he linkage be ween he impo ance, pe o mance,
and cus ome sa is ac ion o ho el se ice a ibu es, acco ding
o Hu e al. (2020). Fu he mo e, in his ega d, Luo e al. (2021)
in oked sen imen analysis in examining economy ho els in Chi-
na and de e mined ho el se ice elemen s ha nega i ely in lu-
ence cus ome expe ience. Ano he no able s udy by Nie e al.
(2020) sugges ed a unique app oach o building a ho el ecom-
menda ion sys em g ounded in blending mul iple c i e ia deci-
sion-making (MCDM), sen imen analysis, and la en Di ichle
alloca ion (LDA). These no ed s udies ha e demons a ed ha
sen imen analysis p o ed a obus and eliable ex mining and
analy ical ool in hospi ali y esea ch o examine and de e mine
he an eceden s o ho el gues sa is ac ion.
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es h ough Deep Lea ning o online e iews 75
3. RESEARCH METHODOLOGY
To add ess he posed RQ and mee his explo a o y s udy's
aims, he de eloped esea ch me hodology comp ised i e se-
quen ial da a collec ion and enginee ing phases, applying a se ies
o da a analysis p ocedu es and isualising he esul s. Figu e 1
depic s he de eloped me hodology and he logic we ollowed in
his esea ch.
3.1. Sampling p ocedu e
This esea ch used a simple andom sampling (SRS) p oce-
du e o es ablish a esea ch se ing o his s udy. I is assumed
ha SRS canno ackle no o ious gene alisa ion issues because
des ina ions, ho el se ice a ibu es, and, hus, gues expe ienc-
es may a y (Malina e al., 2011). Ne e heless, such an app oach
complies wi h he accu acy o he empi ical s udy design in a
single se ing o ge gene alisable cus ome insigh s in o he ho-
el se ices belonging o a pa icula ou is des ina ion (Ma iani
e al., 2019).
Figu e 1
Resea ch me hodology scheme
Sou ce: Own elabo a ion.
The p esen s udy employed a ou -s age SRS p ocedu e. Fo
his pu pose, we i s applied an online andom digi gene a o
(h ps:// andom.o g/) o pick he numbe be ween 5 and 20 ou
o he 100 mos isi ed ou is ci y des ina ions in 2022. One hun-
d ed ci ies we e conside ed because he ou ism analysis li e a-
u e commonly employs his pa icula numbe o op pe o m-
e s in calcula ing he ci y des ina ions index (Popo a, 2023). Also,
i was assumed ha mos isi ed ci ies we e a good sou ce o su i-
cien gues e iew a ailabili y equi ed o analysis. The gene a o
had e u ned 11, so his esea ch collec ed da a om ele en des i-
na ions o be de ined u he by SRS. As Bal es and Ralph (2022)
sugges ed o SRS p ocedu es, we used a andom d aw om he
ci y a ea phone codes o 100 popula des ina ions o de e mine 11
ci ies o da a collec ion (Table 1). Nex , he andom digi gene -
a o picked 4-s a ho els om he 2-5 s a s op ion. We hen op -
ed o booking.com as a cus om da a sou ce pla o m commonly
u ilised in hospi ali y esea ch (Sann e al., 2021).
3.2. Da a collec ion
This esea ch employed Py hon eques s, beau i ulsoup, and
lxml lib a ies o sc ape and pa se he ho el gues e iews and
build a e iew ex co pus o u he analysis. We con igu ed
he c awle o sc ape he signi ican pa s o e iews o o igina e
he a iables and hei espec i e alues equi ed acco ding o
he de eloped esea ch design. The a ge ed da a included ho el
name, coun y o a e iewing gues , ho el a ing sco e gi en by
he e iewe , and e iew ex i sel — i s ‘posi i e’ and ‘nega i e’
pa s, as booking.com spli s he gues e iew o m in o hese
wo ca ego ies. The web c awle was p og ammed o sea ch and
sc ap e iews ele an o ou -s a ho els in all w i en languag-
es in ele en des ina ions. Nex , acco ding o he sugges ions o
Ma iani e al. (2019), he c awle was se o skip blank o incom-
ple e e iews whe e he ex leng h was less han 15 wo ds. The
yea 2022 was a pe iod o be conside ed by he web c awle . A -
e i s un, he c awle sc aped N=109715 ho el gues e iews o
a abula ed da ase om 11 ci y des ina ions de e mined by he
SRS p ocedu e (Table 1).
Table 1
Ci y des ina ions and numbe o sc apped e iews
## Des ina ion Coun y n o e iews
1 Bo deaux F ance 2270
2Dubai OAE 10550
3 Hawaii USA 10352
4 Khu gada Egyp 12902
5 Las Vegas USA 11429
6London UK 11556
7Malaga Spain 10156
8 Munich Ge many 3345
9New Yo k USA 17978
10 P ague Czech Rep. 8880
11 Qa a Qa a 10297
To al n o e iews: 109715
Sou ce: Own elabo a ion.
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
76 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
As he design o his s udy equi ed he ex pa s o he e-
iews solely o comple e sen imen analysis, he coun y o he
e iewe and ho el a ing sco e a iables we e disca ded om
he cu en analysis o he needs o u u e s udies. The ob ained
da a equi ed auxilia y da a enginee ing o ecei e a comple e
da ase eady o analy ical p ocedu es. In his ein, as he e-
sea ch design equi ed he sen imen o he en i e gues e iew
o be gauged, we i s conca ena ed columns in he e ie ed da-
ase con aining a ailable ex pa s in o a single a iable ep e-
sen ing he ull gues e iew (Figu e 2).
Figu e 2
Pa sed da a se agmen
(da ase wi h conca ena ed e iew be o e emo al
o unnecessa y da a columns)
Sou ce: Own elabo a ion.
The mined gues e iews we e w i en in di e en na ional
languages. Whe eas Py hon na u al language p ocessing (NLP)
amewo ks demons a e hei bes capabili ies when applied o
ex s in English, i was essen ial o ansla e e iew ex s in o his
pa icula language. Py hon NLP VADER sen imen analysis li-
b a y, applied in his esea ch and which we depic below in his
sec ion, p o ides such ansla ion op ions. VADER execu es ex
ansla ion o English by u ilising a p e- ained Google Cloud
T ansla e API base wi h a high le el o ansla ion accu acy, ac-
co ding o li e a u e Wang, L., & Ki ilenko, A. P. (2021). On op
o ha , we addi ionally employed wo o he online ansla o s,
IBM Wa son Language T ansla o and Yandex.T ansla e o en-
su e a co ec ansla ion on en andomly selec ed e iews o
e e y language ound in he da ase . This p ocedu e helped e i-
y no loss o meaning, as all es ed ex s had he same deno a ion
in English.
3.3. Tex da a p ep ocessing
Almos all he machine lea ning asks ele an o NLP e-
qui e addi ional p ocedu es o enginee collec ed ex da a o
subsequen analysis. These p ocedu es, a i s , help o spli ex
da a in o smalle chunks, e e ed o as okens, and hen apply
auxilia y echniques o ensu e he p ecision o he esea ch e-
sul s (Pe kins, 2014). This esea ch employed o dina y da a p e-
p ocessing p ocedu es ele an o NLP, as Alam and Yao (2019)
sugges ed. These essen ial NLP p ocedu es can be un by he nl k
( e e ed o as Na u al Language Toolki ) Py hon amewo k,
ecognised as an indus y-s anda d solu ion o ex da a mining
p ocedu es.
We commenced he applica ion o NLP p ep ocessing ech-
niques wi h ex okenisa ion (also known as lexing) o con e
wo ds in o measu emen uni s, o okens, by emo ing punc u-
a ion and whi espaces, making hem eady o analy ical manip-
ula ions. Fo his pu pose, we a ou ed he UPPipe okenise ,
which is no ed as a e sa ile and eliable ool o ex okenisa ion
acco ding o he ex an body o li e a u e (S aka & S ako á,
2017). Fu he mo e, a e okenisa ion, we sequen ially u ilised
h ee p ocedu es, namely, (a) ex ans o ma ion by elimina ing
le e accen s ound in some wo ds; (b) ex no malisa ion whe e
we an s emming gea ed by he UDPipe lemma ise (S aka
e al., 2016); and (c) ex il e ing o disca d egula exp essions
and s op-wo ds om he ex . These p ocedu es we e essen ial
o educe bias and a ain mo e p ecise esul s o analysing he
e iews’ ex da a.
3.4. Da a analysis p ocedu es
This s udy u ilised h ee sequen ial da a analy ical echniques
comp ising sen imen analysis, logis ic eg ession and seman ic
ne wo k analysis (Figu e 1). Sen imen analysis was signi ican
o his esea ch o so ou and zoom in on he nega i e e iews
ha would se e as a sou ce o e eal he o iginal gues s' insigh s
on hei ho el expe iences. These insigh s we e essen ial o dis-
ce n se ice elemen s ha ou inely caused nega i e cus ome
expe iences in he sampled ho els.
Fo sen imen analysis, his s udy has applied VADER
(Valence Awa e Dic iona y o Sen imen Reasoning), a lexi-
con-based solu ion in NLP capable o gauging pola i y and i s a-
lence simul aneously. Thanks o hese ad an ages, he esea che
ecei es mo e p ecise sen imen sco e alues. This is plausible
because he VADER amewo k is no limi ed o elying solely on
a lexicon while making i s compu a ions; i is also a ule-based
applica ion. In summa y, his means ha VADER eckons sen i-
men alence by he con ex o a speci ic wo d while compu ing
i s sen imen and conside ing capi alisa ion and e en emojis i
hey accompany wo ds o a e s andalone in he analysed ex .
These VADER’s capabili ies make i a mo e eliable app oach o
ecei ing p ecise sen imen alues han he o he amewo ks
(Hu o & Gilbe , 2014).
Nex , he subsequen da a analysis s age applied he logis-
ic eg ession echnique. I was necessa y o e eal he mos
impac ul ho el se ice elemen s ha p ecipi a e low gues
sa is ac ion. As logis ic eg ession equi es a bina y depend-
en a iable, we no malised VADER’s con inuous compound
sen imen sco e o 0 and 1 o mee his equi emen . A alue
o «0» ep esen ed a nega i e gues e iew sen imen , while
«1» deno ed a posi i e sen imen in he upda ed da ase . Al-
hough he logis ic eg ession helps highligh he ho el se ice
mic o-elemen s causing bad gues expe iences, i s ill p o ides
a pauci y o in o ma ion o gene a ing p ope insigh s in o
se ice ailu es.
Con e sely, he unco e ed con ex o he ound se ice mi-
c o-elemen s is a p ac ical means o ob ain mo e in o ma ion
on he an eceden s o gues us a ions and disce n he de ails
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es h ough Deep Lea ning o online e iews 77
o ho el se ice ailu es. Fo his eason, his esea ch employed
seman ic ne wo k analysis (Oh & Kim, 2020) o shed mo e
ligh on he con ex o mal unc ioning ho el se ice mic o-el-
emen s. By aiming o ob ain a mo e p ecise comp ehension o
such con ex , we implemen ed seman ic ne wo k analysis sole-
ly on he e iews wi h nega i e sen imen sco es ollowing a
sugges ion om p io li e a u e (Bachleda & Be ada-Fa hi,
2016; Is aeli e al., 2019). Such e iews we e so ed om he
ini ial da ase and pu in o a subsample o n = 23135. The nex
sec ion o his pape depic s he esul s a ained a e applying
seman ic ne wo k analysis and he es o he abo e-explica ed
da a analysis p ocedu es.
4. RESULTS
4.1. Gues e iew sen imen analysis
As no ed ea lie , his s udy has employed he VADER ame-
wo k o implemen sen imen analysis. VADER algo i hm anal-
yses ex and e u ns ou sen imen sco es: nega i e, neu al,
posi i e, and compound. These sco es ha e alues in he ange
o —1 o 1. VADER compu es he compound sen imen alue
by no malising he sum o he o he h ee sco e alues (Hu o &
Gilbe , 2014). The ou comes o he VADER p ocedu e applica-
ion o sen imen analysis a e depic ed in Figu e 3.
Figu e 3
Re iews da ase exce p wi h VADER sen imen analysis ou pu alues.
No e: This igu e ep esen s he i s i e ows o he en i e da ase , demons a ing he ou pu o unning he VADER p ocedu e.
Legend: ‘Ho el’ – ho el o which gues e iew was pos ed; ‘ e iew’ – gues e iew ex displayed pa ially due o he Py hon clien limi a ions;
‘coun y’ – coun y o gues na ionali y; ‘sco e’ – ho el a ing sco e se by gues s; ‘ ade _sco es’ – Py hon a iable o dic iona y ype {‘key’: ‘ alue’}
con aining compu ed sen imen alues o each gues e iew; ‘ ade _compound’ - a no malised sum o VADER’s posi i e, neu al and nega i e
sen imen sco e alue; ‘ ade _pos’ - VADER’s posi i e sen imen sco e alue; ‘ ade _neg’ - VADER’s nega i e sen imen sco e alue; ‘ ade _neu’
- VADER’s neu al sen imen sco e alue.
Sou ce: Own elabo a ion.
As indica ed p e iously in he me hodology sec ion, ob aining
sen imen sco e alues acili a ed a gene a ion o he dependable
bina y a iable, making i possible o implemen logis ic eg ession.
4.2. Logis ic eg ession
The applied logis ic eg ession was se o Ridge (L2) egula -
isa ion ype, cos s eng h C=1, sampling c oss- alida ion wi h
10 olds, and balance class dis ibu ion o e eal posi i e coe -
icien alues o he ho el se ice elemen s ha imply a eason
o a nega i e e iew sen imen . The esul ing logis ic eg ession
model un eiled adequa e model e alua ion s a is ics (Table 2).
Table 2
Logis ic eg ession model e alua ion
Model P ecision Recall* AUC** CA*** F1****
Logis ic Reg ession 0.928 0.960 0.936 0.909 0.943
No e: * — in he domain o E o ype I, his alue deno es he p opo ion
o ue posi i e alues among all posi i e obse a ions o he da ase ;** —
ep esen s he squa e a ea below he p edic ion ROC (Recei e Ope a ing
Cu e, Figu e 4); *** — implies classi ica ion accu acy, e.g., he sha e
o ins ances ha we e adequa ely classi ied; **** — deno es a mean o
p ecision and ecall weigh ed ha monically F1sco e =2(P ecision Recall)
(P ecision +Recall)
.
Sou ce: Own elabo a ion.
ROC analysis is a cogen ool u ilised o e i y he accu acy
o he de eloped model (Fawce , 2006). I compa es he model’s
alse posi i e (FP) a e o speci ici y wi h he maximum p oba-
bili y ha a ge s 1 while he ac ual alue = 0 wi h he model’s
ue posi i e (TP) sensi i i y whe e p obabili y a ge s 1 while
ue alue = 1. Figu e 4 isualises he model e alua ion esul s
using he ROC analysis. Acco ding o Fawce (2006), he model
accu acy is e iden when he cu e is abo e he dashed line ep-
esen ing he non-disc imina o y es and nea he g aph's le
and op bo de s. The ROC cu e complies wi h his equi emen ,
p o iding u he e idence o he high deg ee o he de eloped
model accu acy (Figu e 4).
Logis ic eg ession e u ned 281 okens in he da ase as in-
dependen a iables. Many o hese a iables needed a meaning-
ul sense and hus we e no aluable o u he da a analysis.
To disca d hem, we implemen ed an NLP p ocedu e known as
POS- agging o ex ac nouns om he a iable lis . As a pa
o speech (POS), nouns commonly ep esen he ho el se ice
elemen s in gues e iews (Gee ha e al., 2017). We ex ac ed o-
kens ele an o he noun POS and ecei ed 50 a iables solely
pu suan o he ho el se ice mic o-elemen s ha gene a e gues
esen men wi h he ho el (Annex I).
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
78 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
Figu e 4
Logis ic eg ession e alua ion wi h he ROC analysis
Sou ce: Own elabo a ion.
Finally, we highligh ed he op en independen a iables
ep esen ing ailing se ice mic o-elemen s o sc u inise hei
con ex and analyse he easons o hei o igina ion (Table 3).
Table 3
Top en ho el se ice elemen s p ecipi a ing nega i e e iew
sen imen (logis ic eg ession coe icien (β) alues)
Ta ge a iable alue = 0 (nega i e sen imen sco e)
Reg ession in e cep 0.068 nega i e e iews
coun :*
n = 23135
TF-IDF
sco ing μ:**
Failing se ice
mic o-elemen s β alue
(ai ) condi ioning 0.365 458 0.003
ca pe 0.358 767 0.003
gym 0.356 162 0.001
ke le 0.331 373 0.002
able 0.307 300 0.001
luggage 0.300 902 0.004
pay 0.295 352 0.002
in e ne 0.268 306 0.002
pic u es 0.254 399 0.002
0.228 1019 0.003
* — n o gues e iews con aining he se ice elemen ;
** — deno es a mean o he oken ele an o a ho el se ice mic o-
elemen compu ed wi h TF-IDF (Te m F equency-In e se Documen
F equency) me ic ha is a s a is ical me hod used o de e mine
he signi icance o a wo d conce ning a documen wi hin a se o
documen s.
Sou ce: own elabo a ion
4.3. Seman ic ne wo k analysis
A he inal s age o da a analy ical p ocedu es applica ion,
his esea ch applied ne wo k analysis o he op en poo ly op-
e a ed ho el se ice elemen s e ie ed om logis ic eg ession
(Table 3). Be o e unning seman ic ne wo k analysis, execu ing
an NLP echnique o Conco dance o build a con ex a ound he
oken ep esen ing a speci ic se ice mic oelemen is essen ial.
Conco dance equi es a se ing o N-g am ange, namely he
numbe o okens su ounding an analysed se ice mic oele-
men in e e y gues e iew whe e i can be ound. This s udy
employed an N-g am ange o 6, meaning h ee okens be o e
and h ee okens a e he se ice mic oelemen , o gene a e am-
ple con ex a ound i o be analysed u he .
The in-dep h sc u iny o he seman ic ne wo k analysis ap-
plied o he p incipal dys unc ional ho el se ice mic o-elemen s
e eals se e al indings. A i s , he logis ic eg ession applica-
ion indica ed ha (ai ) condi ioning (β = 0.365) is he op se -
ice mic oelemen which de e io a es gues sa is ac ion. The
ne wo k map (Annex II, Figu e 1) di ulges he de ails accompa-
nying gues us a ions, including no ai condi ioning a ailabili-
y in ho el ooms and di y, noisy, b oken o imp ope ly wo king
appliances. Second, ho el and hallway oom ca pe s (β = 0.358)
ecei ed many gues complain s, acco ding o ou indings (An-
nex II, Figu e 2). Gues s poin ed o he old, s ained, di y, wo n,
smelly ca pe ing in hei ho el e iews. Thi d, ho el gues s we e
dissa is ied wi h he gym acili ies (β = 0.356). In his line, small,
di y, poo ly equipped acili ies and he swimming pool a e ea-
sons o gues s' i i a ion (Annex II, Figu e 3). Nex , he ke le
was anked he ou h mic oelemen con ibu ing o he nega-
i e pe cep ion o he ho el se ices (β = 0.331). Seman ic maps
poin o he ke le's una ailabili y in he oom, aul y o b oken
appliances, and limi ed o no a ailabili y o cups, mugs, and
ea sache s (Annex II, Figu e 4). A ho el oom u ni u e piece
such as a able was he i h op complain no ed in ho el e iews
(β = 0.307). In his ega d, gues s highligh ed he absence o a-
bles o chai s and men ioned di ound on his u ni u e i em
(Annex II, Figu e 5).
Fu he mo e, he ou comes o his s udy poin o luggage
as he six h op- anked mic oelemen , p edisposing gues dis-
appoin men owa d ho el se ices (β = 0.300). Online e iews
wi h a nega i e sen imen indica e ha gues s may expe ience
di icul ies wi h luggage s o age se ice a he ecep ion and li -
le help om he s a o ca y hea y luggage o he ho el oom
(Annex II, Figu e 6). Conce ning paymen (β = 0.295), gues s'
dissa is ac ion a ise om ex a paymen s, need o money de-
posi s, high p icing o he supplemen a y se ices and inco ec -
ly cha ged bank ca ds (Annex II, Figu e 7). Paymen is ollowed
by in e ne (β = 0.268) which gene a es a educ ion in gues sa -
is ac ion because o wi- i una ailabili y o poo , slow, and inad-
equa e connec ion in he ho el oom (Annex II, Figu e 8). In e -
es ingly, gues s poin ed o one i em which is no di ec ly linked
o he ho el p oduc consump ion bu pe ains o he ini ial s age
o he cus ome jou ney in hospi ali y. This i em is ele an o
pic u es (β = 0.254) ha ho els employ in hei online ma ke ing.
Conce ning pic u es, ho el gues s massi ely no ed a disc epancy
be ween he online ho el pho os and he ac ual iews hey see in
he ho el (Annex II, Figu e 9). Finally, a se ice mic oelemen
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es h ough Deep Lea ning o online e iews 79
o TV eme ged as he en h op eason o gues s’ esen men
exp essed in hei e iews (β = 0.228). In his domain, gues s
ou inely complained o non-wo king, b oken o una ailable e-
mo e con ol, small sc een size, old TV se s and limi ed channel
choices (Annex II, Figu e 10).
5. Discussion
By employing sen imen and seman ic ne wo k analyses on
109715 ho el e iews wi h he applica ion o machine lea ning
echniques o big da a om online e iews collec ed om ele -
en ci y des ina ions, his s udy has exposed and dis inguished
he causes o low gues sa is ac ion. Consequen ly, his esea ch
has de e mined i y ho el mic o-se ice elemen s ha may com-
monly induce ho el se ice ailu es (Annex I). Simul aneously,
a p io mains eam body o esea ch has ended o illumina e
g oups o ho el se ice elemen s o ac o s a he han pa icu-
la se ice elemen s. In line wi h he p io s udies, his esea ch
highligh s speci ic ho el-gues ouch poin s and de e mines ho el
se ice a ibu es equi ing immedia e manage ial ocus.
In his ein, ou s udy con i ms he p io esea ch indings,
which e ealed signi ican mic o-elemen s o possible se ice
ailu es a ising om gues s’ low sa is ac ion in he domain o
he ho el oom ac o . These mic o-elemen s comp ise ai condi-
ioning, bed, noise, owels (Luo e al., 2021), wi i (in e ne ) signal
quali y (Hu e al., 2021), cleanness (Za ezadeh e al., 2022), and
hai d ye (Pa k e al., 2019). Howe e , ou esea ch goes beyond
he ex an co pus o s udies as i has e ealed a b oade ange o
se ice mic o-elemen s ele an o se ice ailu es unde he ho-
el oom ac o . Complimen a y o p io li e a u e, ailing se ice
mic o-elemen s de e mined by his s udy comp ise oom ca pe ,
ke le, able, TV, walls, ma ess, oom space, showe , ub, sink,
smell, idge, u ni u e and balcony. In e es ingly, he indings
o ou s udy poin o he ho el oom as he key ac o in gues
sa is ac ion educ ion and, hus, he mos equen a ea o ho-
el se ice ailu es, as 28 ou o 50 se ice mic o-elemen s a e
pe inen o he ho el oom. In addi ion, applying he seman ic
ne wo k analysis has made i possible o del e in o he con ex
o he ho el se ice mic o-elemen s. I helped o e eal deepe
unde lying easons and oo s o gues us a ion wi h pa icula
ho el se ice mic o-elemen s.
Simila ly o La inopoulos (2020), he comple ed s udy con-
i med he signi icance o he ho el loca ion and ex e io as a
ac o o gues sa is ac ion. On op o ha , ou s udy has e ealed
addi ional mic o-elemen s ele an o his ac o , namely, a ea,
pa king, and (ho el) buildings. In he ealm o ho el acili ies
ac o , whe eas ex an esea ch has highligh ed solely he impac
o hei o e all unc ionali y on pe cei ed se ice ailu es (Ying
e al., 2020), his s udy conc e ely poin s o he ho el gym as a
mic o-elemen o se ice ailu e.
In ea lie s udies, he ole o ho el pe sonnel in con ibu -
ing o ho el se ice ailu es has been widely acknowledged (Nie
e al., 2020). Building upon hese esea ch indings, his s udy
del es deepe in o iden i ying speci ic ac i i ies ha highligh
a eas whe e gues s o en expe ience us a ions ela ed o ho-
el pe sonnel. These a eas include luggage handling, ecep ion
se ice, shu le bus se ice, and anima ion in eso ho els.
Fu he mo e, ou s udy aligns wi h he indings o he s udies,
which highligh ed he signi icance o ood and be e age (F&B)
in in luencing ho el gues sa is ac ion (Nie e al., 2020; Philips
e al., 2017). Su p isingly, ou esea ch e eals a ela i ely lowe
signi icance o ho el se ice mic o-elemen s ela ed o F&B in
his espec . Acco ding o ou indings, only ou ele an i ems,
namely ' ood', 'd inks', 'b eak as ', and ' es au an ', we e iden i-
ied as con ibu ing o se ice ailu es. No ably, hese i ems we e
kep om he op en mic o-elemen s lis , and hei posi ions,
based on he log eg ession coe icien alues, we e 16, 26, 34,
and 39, espec i ely, in he whole 50 mic o-elemen s ange.
5.1. Theo e ical implica ions
The comple ed s udy con ibu es o he ho el ma ke ing he-
o y in se e al ways. Fi s , g ounded on he cus ome ’s insigh s
om he ho el e iews, his esea ch has zoomed in on mal unc-
ioning ho el se ices and de e mined singula se ice mic o-el-
emen s ep esen ing se ice ailu es ha cause nega i e cus ome
expe iences and low gues sa is ac ion. Ou esea ch has ex end-
ed and sys emised a se o se ice mic o-elemen s p e iously
sca e ed among a ious s udies. We posi ha highligh ing ho el
se ice ac o s, o ‘a ibu es’, in gues expe ience esea ch, a he
han mic o-elemen s, can lead o a biased unde s anding o he
eal easons o ho el gues dissa is ac ion. As p e iously no ed,
each ho el se ice ac o ypically amalgama es se e al se ice
mic o-elemen s. Each o hem is mo e o less signi ican o he
ho el gues expe ience. Hence, we asse ha zooming in on he
mic o-elemen s is way mo e signi ican o he hospi ali y ma -
ke ing esea ch han doing so on he ho el se ice mac o ac o s.
Second, building upon he abo e poin , his s udy p o ides
a deepe unde s anding and expands he ange o ho el se ice
mic o-elemen s iden i ied in p e ious li e a u e. In alignmen
wi h p io esea ch, ou indings co obo a e ha he in e ne
o Wi-Fi connec ion is among he op mic o-elemen s ha gen-
e a e nega i e gues expe iences and subsequen dissa is ac ion.
Addi ionally, ou s udy highligh s he signi ican ole o ai con-
di ioning in shaping gues s' pe cep ions o ho el se ice quali y.
No ably, Wi-Fi connec ion and ai condi ioning eme ge as con-
sis en op en mic o-elemen s in he p esen esea ch and p e-
ious s udies.
Mo eo e , al hough ou indings align wi h p io esea ch
ega ding he less signi ican se ice mic o-elemen s anked
11-50 (Table A.1.1.), his s udy iden i ies addi ional se ice mi-
c o-elemen s ha play a c ucial ole in shaping gues dissa is-
ac ion wi h he ho el and signi ican ly con ibu e o he o e all
unde s anding o ho el se ice ailu es. Ou s udy subs an ially
con ibu es o he exis ing li e a u e by unco e ing hese p e i-
ously undocumen ed mic o-elemen s. In summa y, his esea ch
imp o es he unde s anding o he mix o ho el se ice mic o-el-
emen s and expands i s scope by cap u ing o e looked ac o s
and highligh ing hei impac on gues sa is ac ion. Including
hese no el indings adds aluable insigh s o he li e a u e and
deepens ou comp ehension o he mul i ace ed na u e o ho el
se ice ailu es.
Thi dly, his s udy ex ends he unde s anding o gues dissa -
is ac ion wi h ho el se ice quali y by shi ing he ocus owa ds
explo ing se ice mic o-elemen s in ele en di e se ci y des ina-
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
86 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
APPENDIX A1.2.
Seman ic ne wo k maps o op en mic o-elemen s causing ho el se ice ailu es
Figu e A.1.2.1
Ne wo k map: Ai condi ioning
Figu e A.1.2.2.
Ne wo k map: Ca pe
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
Imp o ing gues sa is ac ion by iden i ying ho el se ice mic o-elemen s ailu es h ough Deep Lea ning o online e iews 87
Figu e A.1.2.3.
Ne wo k map: Gym
Figu e A.1.2.5.
Ne wo k map: Table
Figu e A.1.2.4.
Ne wo k map: Ke le
Figu e A.1.2.6.
Ne wo k map: Luggage
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
88 Se gey Kazako , Ped o Cues a-Valiño, Ve a Bu kouskaya, Daniel Mu a sky
Figu e A.1.2.7.
Ne wo k map: Pay
Figu e A.1.2.9.
Ne wo k map: Pic u es
Figu e A.1.2.8.
Ne wo k map: In e ne
Figu e A.1.2.10.
Ne wo k map: TV
Sou ce o all igu es in Appendix A1.2:Own elabo a ion.
Managemen Le e s / Cuade nos de Ges ión 25/1 (2025) 71-88
