An on ology-based da a in eg a ion app oach o web analy ics in e-comme ce
Ma ía del Ma Roldán Ga cía, José Ga cía-Nie o
∗, José F. Aldana-Mon es
1
Dep . de Lenguajes y Ciencias de la Compu ación, Uni e si y o Málaga, ETSI In o má ica, Campus de Tea inos, Málaga - 29071, Spain
Keywo ds:
Seman ic model
On ology
E-comme ce
Web analy ics
a b s a c
Web analy ics has eme ged as one o he mos impo an ac i i ies in e-comme ce, since i allows com-
panies and e-me chan s o ack he beha io o cus ome s when isi ing hei web si es. The e exis a
se ies o ools o web analy ics ha a e used no only o acking and measu ing web affic, bu also
o analyzing he comme cial ac i i y. Howe e , mos o hese ools ocus on low le el web a ibu es and
me ics, making o he sophis ica ed unc ionali ies and analyses only a ailable o comme cial (non- ee)
e sions.
In his con ex , he SME-Ecompass Eu opean ini ia i e aims a p o iding e-comme ce SMEs wi h ac-
cessible ools o high le el web analy ics. These so wa e acili ies should use di e en sou ces o da a
coming om digi al oo p in s alloca ed in e-shops, o use hem oge he in a cohe en way, and o
make hem a ailable o ad anced da a mining p ocedu es. This mo i a ed us o p opose in his wo k an
on ology-based app oach o collec , in eg a e and s o e web analy ics da a, om many sou ces o popula
and comme cial digi al oo p in s. As a icle’s main impac , we ob ain en iched and seman ically anno-
a ed da a ha is used o p ope ly ain an in elligen sys em, in ol ing da a mining p ocedu es, o he
analysis o cus ome beha io in eal e-comme ce si es. In conc e e, o he alida ion o ou seman ic ap-
p oach, we ha e cap u ed and in eg a ed da a om Google Analy ics and Piwik digi al oo p in s alloca ed
in 15 e-shops o di e en comme cial sec o s and coun ies (UK, Spain, G eece and Ge many), h ough-
ou se e al mon hs o ac i i y. The ob ained esul s show di e en pe spec i es in cus ome ’s beha io
analysis ha go one s ep beyond he mos popula web analy ics ools in he cu en ma ke .
1. In oduc ion
In he las ew yea s, web analy ics has eme ged as one o he
mos impo an ac i i ies in e-comme ce, since i allows companies
and e-me chan s o ack he beha io o cus ome s when isi ing
hei e-shop si es. Web analy ic applica ions can also help compa-
nies o measu e he esul s o adi ional p in o b oadcas ad e -
ising campaigns. Web analy ics p ocedu e is based on measu ing
a isi o ’s beha io once on a gi en e-shop si e, which includes i s
d i e s and con e sions ( o ac ual cus ome ). These da a a e ypi-
cally compa ed agains key pe o mance indica o s and used o im-
p o e a websi e o ma ke ing campaign’s audience esponse.
∗Co esponding au ho .
E-mail add esses: [email p o ec ed] (M.d.M.R. Ga cía), [email p o ec ed] ,
In he cu en ma ke , he e exis a se ies o ools o web an-
aly ics, such as: Google Analy ics, Piwik, Clicky, and S a Coun e ;
ha a e widely used no only o acking and measu ing web a -
fic, bu also o analyzing he comme cial ac i i y, hence o im-
p o e he e ec i eness o a websi e. Howe e , hese ools o en o-
cus on low le el and limi ed se s o web me ics and a ibu es,
wi hou he possibili y o p o iding specialized analyses. In mos o
cases, high le el web me ics and sophis ica ed unc ionali ies a e
a ailable only o comme cial (non- ee) e sions, which a e a ely
accessible by SMEs o indi idual e-me chan s.
In his con ex , he SME-Ecompass Eu opean ini ia i e
2 aims a
p o iding e-comme ce SMEs wi h accessible ools o high le el
web analy ics. These so wa e acili ies use he di e en sou ces o
da a coming om di e en digi al oo p in s alloca ed in e-shops.
Howe e , in eg a ing da a om mul iple he e ogeneous sou ces
en ails dealing wi h di e en da a models, schema and que y lan-
guages. The e o e, he e is a clea demand o in eg a i e p oce-
2 SME-Ecompass FP7 Eu opean ini ia i e h p://www.sme-ecompass.eu/
jmga cianie[email p o ec ed] (J. Ga cía-Nie o), j [email p o ec ed] (J.F. Aldana-Mon es).
1
This wo k is pa ially unded by FP7 EU p ojec SME E-COMPASS unde G an
No: 315637. I is also pa ially unded by G an s TIN2014-58304 (Spanish Minis y
o Sciences and Inno a ion) and Regional p ojec s P11-TIC-7529/P12-TIC-1519. Au-
ho s hanks o in ol ed e-shops o kindly o e web acking da a o es ing and
alida ion.
du es o p o iding he ad anced da a mining algo i hms wi h a
uni o m access o mul iple he e ogeneous web da a sou ces.
The main hypo hesis in his wo k is: (H1) an on ology-based
in eg a ion app oach will help us o collec , use he da a o-
ge he in a cohe en way, and s o e web analy ics da a, om
many sou ces o popula and comme cial digi al oo p in s . As a
esul , (H2) we will ob ain en iched and seman ically anno a ed
da a ha will be able o ain da a mining p ocedu es o ad-
anced analysis o cus ome beha io in eal e-comme ce si es .
This mo i a ed us o p opose a seman ic app oach ha uses an
on ology as a media ed schema o he ep esen a ion and consoli-
da ion in a knowledge base o he acking da a om web sou ce’s
seman ics. Seman ic web on ologies become a key echnology o
in elligen knowledge p ocessing, p o iding a amewo k o sha -
ing concep ual models abou a domain. Seman ic mappings be-
ween he sou ce schema and he on ology a e hen defined and
used o ans o m he o iginal da a o RDF (Resou ce Desc ip-
ion F amewo k)
3
. This way, da a om he e ogeneous sou ces a e
s o ed and in eg a ed inside a single RDF eposi o y, which can be
now easily que ied by high le el algo i hms. The goal is o p op-
e ly eed a ificial in elligence p ocedu es capable o deciding how
o pe o m ma ke ing ac i i ies, such as: displaying a gi en ad e -
isemen a ge ed o ce ain ca ego y o clien s, o dec easing he
p ice o a p oduc in a gi en egion; hen gi ing ise o sophis i-
ca ed expe sys ems o e-comme ce applica ions.
The main con ibu ions o his s udy a e summa ized as ol-
lows:
–We ha e de eloped a seman ic app oach o he da a in eg a-
ion and consolida ion o mul iple web analy ics da a sou ces.
These da a a e daily accumula ed om many he e ogeneous
digi al oo p in s alloca ed on ac ual e-shops.
–We
ha e designed and implemen ed o he fi s ime an OWL
(Web On ology Language) on ology ( Dean & Sch eibe , 2004 )
o web analy ics. This on ology conside s a la ge and comple-
men ed se o a ibu es and me ics, which ha e been oken
om se e al ep esen a i e web analy ics ools in he ma ke .
–To
es hypo hesis H1, we ha e cap u ed and in eg a ed da a
om Google Analy ics and Piwik digi al oo p in s alloca ed in
15 e-shops o di e en comme cial sec o s ( e ail, ou ism, elec-
onics, pha macy, e c.) and coun ies (UK, Spain, G eece and
Ge many), h oughou se e al mon hs o ac i i y. The da a a e
in eg a ed ollowing he same (s anda d) o ma and s o ed in
a common RDF eposi o y.
–To
es hypo hesis H2, ob ained “seman ized” da a a e used o
ain ad anced da a mining algo i hms o pe o m cus ome ’s
p ofile analyses. In pa icula , hese algo i hms a e es ed wi h
success in wo cases o s udy o classi y he isi o ’s beha io
and p oduc p e e ence.
The emaining o his a icle is o ganized as ollows. In
Sec ion 2 , backg ound and li e a u e o e iew a e p esen ed.
Sec ion 3 p esen s he cu en s a e and p ac ices in web analy ics
o e-comme ce. In Sec ion 4 , he seman ic app oach is desc ibed,
gi ing de ails o he se ice a chi ec u e and he OWL on ology. A -
e his, he alida ion p ocedu e is epo ed in Sec ion 5 . Finally,
main conclusions and u u e wo k a e gi en in Sec ion 6 .
2. Backg ound and ela ed wo k
This sec ion desc ibes he main backg ound concep s. A e iew
o cu en ela ed wo ks in he specialized li e a u e is ca ied ou
o poin ou hei main di e ences wi h ega ds o ou app oach.
3 RDF in W3C h ps://www.w3.o g/RDF/
2.1. Backg ound concep s
- On ology. On ologies p o ide a o mal ep esen a ion o he
eal wo ld, sha ed by a sufficien amoun o use s, by defining con-
cep s and ela ionships be ween hem ( G ube , 1993 ). In compu e
and in o ma ion sciences, an on ology defines a se o ep esen-
a ional p imi i es wi h which o model a domain o knowledge
o discou se. These p imi i es a e ypically concep s (classes), a -
ibu es (p ope ies), class membe s (class ins ances) and ela ion-
ships (p ope y ins ances). The defini ions o he ep esen a ional
p imi i es include in o ma ion abou hei meaning and cons ain s
on hei logically consis en applica ion.
On ologies a e pa o he W3C s anda ds s ack o he seman ic
web, in which hey a e used o speci y s anda d concep ual ocab-
ula ies in which o exchange da a be ween sys ems, p o ide se -
ices o answe ing que ies, publish eusable knowledge bases, and
o e se ices o acili a e in e ope abili y ac oss mul iple, he e o-
geneous sys ems and da abases.
-RDF. Resou ce Desc ip ion F amewo k is a basic on ology lan-
guage used o ep esen ing in o ma ion abou esou ces on he
web ( S aab & S ude , 2009 ). Resou ces a e desc ibed in e ms o
p ope ies and p ope y alues using RDF s a emen s. S a emen s
a e ep esen ed as iples, consis ing o a subjec , p edica e and
objec . RDF Schema ( S aab & S ude , 2009 ) (RDFS) “seman ically
ex ends” RDF o enable us o alk abou classes o esou ces, and
he p ope ies ha will be used wi h hem. I does his by gi -
ing pa icula meanings o ce ain RDF p ope ies and esou ces.
RDFS p o ides he means o desc ibe applica ion specific RDF o-
cabula ies. RDF and RDFS p o ide basic capabili ies o desc ibing
ocabula ies ha desc ibe esou ces, me ada a and on ologies.
-SPARQL. I is an RDF que y language o on ology models
and da abases, capable o ex ac ing and manipula ing in o ma-
ion s o ed in RDF o ma . Essen ially, SPARQL is a g aph-ma ching
que y language ha can be used o ex ac knowledge om he
model such as he one p oposed in his a icle. Gi en a da a sou ce
D, a que y consis s o a pa e n, which is ma ched agains D. The
combina ions o alues esul ing om his ma ching cons i u e he
esul o he que y ( Pé ez, A enas, & Gu ie ez, 2009 ). SPARQL has
s ong suppo o que ying semi-s uc u ed and agged da a, e.g.
da a wi h an unp edic able and un eliable s uc u e. SPARQL sup-
po s que ies o ne wo ked, web da a sou ces iden ified by URIs. In
ac , i is a W3C ecommenda ion o RDF da a.
-OWL. In 2004, he W3C on ology wo king g oup ( Dean &
Sch eibe , 2004 ) p oposed OWL as a seman ic ma kup language o
publishing and sha ing on ologies on he Wo ld Wide Web. F om
a o mal poin o iew, OWL is equi alen o a e y exp essi e de-
sc ip ion logic whe e an on ology co esponds o a Tbox ( G ube ,
1993 ). This equi alence allows he language o exploi desc ip-
ion logic esea che esul s. OWL ex ends RDF and RDFS. When
compa ed o RDF models, OWL adds mo e ocabula y o desc ib-
ing p ope ies and classes: ela ions be ween classes (e.g. disjoin -
edness), ca dinali y (e.g. “exac ly one”), equali y, iche yping o
p ope ies, cha ac e is ics o p ope ies (e.g. symme y), and enu-
me a ed classes ( McGuinness & Ha melen, 2004 ).
-OWL-DL. Syn ac ic a ian o he SHOIN (D) desc ip ion logic
( Haase & S ojano ic, 2005 ) wi h a di e en e minology o OWL,
which is based on RDFS, hence he suppo o da a alues, da a
ypes and da a ype p ope ies. OWL-DL es ic s OWL in o wo
dis inc ways ( Ho ocks & Pa el-Schneide , 2003 ): fi s , some syn-
ac ic cons uc s like ecu si e desc ip ions in hem a e no al-
lowed; second, classes, indi iduals and p ope ies ( espec i ely
concep s, indi iduals and oles in desc ip ion logics) mus all
be disjoin . In his wo k, we use OWL-DL syn ax o o malize
he p oposed on ology he e o ou seman ic model. A summa-
ized desc ip ion o basic OWL-DL seman ics syn ax is shown
in Table 1 , whe e an in o mal logic syn ax is ep esen ed (le
Table 1
Basic OWL-DL seman ic syn ax used o o mally define he p oposed
on ology.
Desc ip ions Abs ac syn ax DL syn ax
Ope a o s in e sec ion ( C
1
, C
2
, , C
n
) C
1
C
2
C
n
union ( C
1
, C
2
, , C
n
) C
1
C
2
C
n
Res ic ions o a leas 1 alue V om C∃ V.C
o all alues V om C∀ V.C
R is symme ic R ≡R
−
Class Axioms A pa ial ( C
1
, C
2
, , C
n
) A  C
1
C
2
C
n
A comple e ( C
1
, C
2
, , C
n
) A ≡C
1
C
2
C
n
column) wi h ega ds o he co esponding OWL-DL equi alen
( igh ).
2.2. Rela ed wo ks
In he las decade, a se ies o s udies ha e been appea ing ha
seman ically model ce ain domains o sub-domains o knowledge
in he con ex o e-comme ce.
A fi s a emp was p oposed in T as ou , Ba olini, and P eis
(2003) , whe e a se ice desc ip ion language was defined o be
used h oughou he li e-cycle o a business- o-business (B2B) e-
comme ce in e ac ion. In pa icula , hey ocused on DAML+OIL, as
i is a sufficien ly exp essi e and flexible se ice desc ip ion lan-
guage o be used no only in ad e isemen s, bu also in ma ch-
making que ies, nego ia ion p oposals and ag eemen s.
A e his, Tamma e al. ( Tamma, Phelps, Dickinson, &
Woold idge, 2005 ) designed an app oach o nego ia ion ac i i ies
in e-comme ce si es. In his wo k, he nego ia ion p o ocol does
no need o be ha d-coded in agen s, bu i is ep esen ed by an
on ology, in e ms o an explici and decla a i e ep esen a ion o
he nego ia ion p o ocol. The on ology is also used o une agen s
s a egies o he specific p o ocol used.
A special case o e-comme ce si es is e- ou ism, o which Wa -
alak ( Wa alak , 2008 ) discussed some on ological ends ha sup-
po he g owing domain o online ou ism. This s udy also ga e
some example concep s o exis ing e- ou ism using on ologies dis-
play in g aphical model and showed hei desc ip ions in OWL and
RDFS syn ax.
Hepp defined wo ela ed on ologies ( Hepp, 2008 ): GoodRela-
ions and P oduc On ology. GoodRela ions is a s anda dized o-
cabula y o e-comme ce (p oduc , p ice, s o e, and company da a)
and he P oduc On ology is an on ology o desc ibing p oduc
ypes based on Wikipedia.
Mo e ecen ly, Ga chalee e al. ( Ga chalee, Li, & Supni hi, 2013 )
p oposed an on ology app oach o co e he knowledge abou
he con en and a chi ec u e o SMEs e-comme ce websi es. This
knowledge is hen used as inpu o a ecommenda ion sys em o
web design, ha cen e ed on he s uc u e o e-shops in Thailand
as sample g oup.
Finally, Akanbi ( Akanbi, 2014 ) p oposed LB2CO, an on ology
which combines he amewo k o IDEF5 & SNAP as an analysis
ool, o au oma ed ecommenda ion o p oduc and se ices. This
on ology is used o model a seman ic amewo k o B2C ansac-
ions ac oss di e en business domains ha acili a es he in e op-
e abili y and in eg a ion o ansac ions o e he web.
As summa ized in Table 2 , all hese wo ks p oposed seman ic
models ocusing on di e en aspec s in he domain o e-comme ce,
such as: web con en s, s uc u e, and li e-cycle ac i i ies. Howe e ,
o he bes o ou knowledge, he e is s ill a lack o wo ks whe e
a seman ic model is used o conside web analy ic a ibu es and
me ics om mul iple and he e ogeneous sou ces o da a. This is a
c i ical issue o cu en e-me chan s ha we y o cope, o he
fi s ime, wi h ou app oach.
In con as o o he pas p oposals, we alida e ou seman-
ic model wi h eal da a coming om digi al oo p in s and web
sc aping me hods in a numbe o ac ual e-shops. As a esul , some
o hese e-comme ce SMEs ha e pu in p ac ice he gene a ed da a
and analyses, leading hem o change and imp o e hei comme -
cial s a egies.
3. E-comme ce web analy ics: cu en p ac ices
P e iously o desc ibe ou seman ic app oach, we summa ize
in his sec ion a se ies o ac i i ies we ca ied ou in he con ex
o SME-Ecompass p ojec , wi h he aim o shedding ligh on he
ac ual s a e o e-comme ce companies.
In a fi s phase, we deli e ed an online su ey o e-shop’s own-
e s (o associa ed chambe s o SME-Ecompass p ojec ) wi h di e -
en ques ions wi h ega ds o: hei company’s p ofile, comme ce
ac i i y, cu en p ac ices when analyzing cus ome ’s and compe i-
o ’s beha io s, e c. A e ha , ace- o- ace in e iews we e also
conduc ed wi h a selec ion o e-me chan s in o de o ob ain de-
ailed in o ma ion o hei p o essional expe ience in e-comme ce.
A comple e epo o hese ques ionnai es wi h he analysis o e-
sponses, s a is ics and conclusions can be ound in Ga ía-Nie o and
Roldán (2014) . In conc e e, mo e han 150 e-comme ce SMEs com-
ple ed he online su eys and 20 ou o hem we e in e iewed in
p i a e sessions. The ollowing conclusions we e ex ac ed:
– Mos o s udied companies a e mic o en e p ises wi h 1 o 5
employees and wo k in Business o Consume (B2C) e ailing
sec o . Mos o hem ha e a maximum numbe o 50 0 0 o de s
pe yea (2013/2014) and a maximum annual e enue o 10,0 0 0
eu os om online sales. The e o e, hey a e a ge candida es
o be beneficia y o au oma ic ( ee o non-expensi e) ools o
ob ain ad anced e-comme ce analysis.
– Conce ning isi o ’s beha io analyses (see Fig. 1 le ), 47% o
companies do no use any ool o hese kind o asks. On he
con a y, a pe cen age o 29% decla ed ha hey use au oma ic
online ools and 22% make hese asks manually. O cou se,
mos o hem ( > 80%) decla ed o be qui e in e es ed on using
a se ice o disco e endencies and common habi s in clien s.
– In e es ingly, as shown in Fig. 1 ( igh ), i can be s ood ou ha
Google Analy ics is he mos used ool o in e iewed compa-
nies (68%), al hough hey also use addi ional ools like Piwik
(16%) and o he (16%). The e o e, he se o common me ics e-
me chan s usually analyze a e hose compu ed by Google Ana-
ly ics, e. g., numbe o isi s, a e age isi ime, geo-localiza ion,
coun y, clien de ice, e c. These me ics a e usually checked
weekly o mon hly.
In he ligh o hese esul s, we decided o ocus ou seman-
ic model on a ibu es and me ics p o ided by Google Analy ics
and Piwik. We selec ed he o me o being he mos used an-
aly ic ool in he ma ke . Howe e , Google Analy ics e-comme ce
ad anced unc ionali y is ( o he da e) no a ailable o ee use s.
This issue led us o complemen ou se o a ibu es wi h hose
o Piwik, since his las ool o e s ee access o ad anced e-
comme ce a ibu es. Addi ional me ics o compe i o and p ice
moni o a e also conside ed in his wo k, which a e ga he ed om
specific web sc apping p ocesses o SME E-Compass ool. I is
wo h no ing ha he p oposed on ology is aimed a co e ing as
gene al as possible web analy ic a ibu es, hence enabling he in-
co po a ion o new analy ics ools in ou seman ic app oach.
4. Seman ic app oach
One o he main aims o his wo k is o cap u e, clean, consol-
ida e and in eg a e da a om di e en sou ces o web acking in
Table 2
Rela ed app oaches in he s a e o he a . The a ge a ea o applica ion, he used on ology/ ocabula y and he pos -
p ocessing analysis, and he alida ion p ocedu e a e epo ed o each wo k.
App oach Ta ge a ea On ology/ ocabula y Analysis Valida ion
T as ou e al. (2003) Li e-cycle in B2B DAML+OIL -No
Tamma e al. (2005) Nego ia ion in e-comme ce Nego ia ion p o ocol Agen s sys em No
Wa alak (2008) E- ou ism E- ou ism concep s Recommende No
Hepp (2008) P oduc s in e-comme ce GoodRela ions -Academic
Ga chalee e al. (2013) E-comme ce si es design Websi e design s uc u e Recommende Academic
Akanbi (2014) B2C ansac ions LB2CO Recommende Academic
P oposal Web analy ics digi al Web analy ics Da a mining Real
oo p in s (Google/Piwik) On ology (WAO) Visi o /p oduc wo ld
Fig. 1. Cu en p ac ices in su eyed e-comme ce SMEs wi h ega ds o he use o au oma ic ools o web analy ics.
e-comme ce si es. Fo his eason, we op ed o design a seman-
ic app oach o sha ing and econcilia ion, whe eby an ag eed on-
ology model is used o a chi e a common unde s anding o he
domain in which he sys em ope a es. In conc e e, we ha e de el-
oped an OWL on ology o desc ibe he e-shops main ea u es by
ollowing he s anda d on ology 101 de elopmen p ocess ( Na alya,
McGuinness, & Debo ah, 2001 ) o se en s eps:
(i) De e mine he domain and scope o he on ology . As he s a -
ing poin , o limi he scope o he on ology, we selec ed
he kind o a iables ha he da a mining algo i hms need
om Google Analy ics and Piwik and also om compe i o s
e-shops, o ins ance: isi o s o igin, isi o s a ibu es, pu -
chasing beha io , p oduc and cus ome de ails, e c.
(ii) Conside eusing exis ing on ologies . As we examined in
Sec ion 2.2 , he e a e no simila on ologies ha ha e been
p e iously p oposed o modeling web acking da a in e-
comme ce. Howe e , we pa ially conside ed wo ela ed on-
ologies: GoodRela ions ( Hepp, 2008 ), which is a s anda d-
ized ocabula y o e-comme ce and he P oduc On ology
( Hepp, 2008 ), which con ex ualizes p oduc ypes based on
Wikipedia.
(iii) Enume a e impo an e ms in he on ology . Impo an e ms
in he on ology we e ex ac ed in a p e ious phase o e-
qui emen s specifica ion ( Ga ía-Nie o & Roldán, 2014 ) om
he minimum se o a iables ha a e needed. Exam-
ples o such e ms a e: add ess, isi o , cus ome , de ice,
b owse , Geog aphical_o igin, Numbe _o _ isi o s, Con e -
sion_ a e, e c.
(i ) Define classes and he class hie a chy . F om he lis o
e ms, we ob ained he on ology classes. Fig. 2 shows he
main se o classes in he hie a chy s a ing om he op
class Thing (  ). These main classes a e ela ed o o he
classes and some o hem ha e subclasses. Fo ins ance,
Analy ic_pa ame e s has a se ies o subclasses, such as:
Bounce_ a e, To al_ e enue, Numbe _o _ e u ning_ isi o s , and
Numbe _o _ ansac ions .
( ) Define he p ope ies o classes and slo s . In o de o ela e
classes and o define a ibu es, we iden ified objec s and
da a ype p ope ies based on he minimum se o a iables
p e iously defined. Examples o objec p ope ies a e: an
e-shop owne is owne o an e-shop, a isi o makes isi s, a
de ice has a b owse , an IP add ess belongs o an o ganiza-
ion, e c. Examples o da a ype p ope ies a e he i le and
URL o a page, he fi s and las name o an e-shop owne ,
he e sion o he ope a ing sys em, he du a ion o a isi ,
e c. An objec p ope y is defined o each subclass o es ab-
lish he co ec ela ionship. Fo example, Page is ela ed o
Bounce_ a e and Da e_o _las _ isi ; E-shop is ela ed o Num-
be _o _cus ome s . Tables 3–8 , desc ibe in OWL-DL ep esen-
a i e subse s o objec and da a p ope ies o a selec ion o
he main classes.
( i) Define he ace s o he slo s . This s ep includes he defini-
ion o ca dinali y cons ain s and alue es ic ions. Value
es ic ions a e used in ou on ology o speci y he da a
ype o he alue in each subclass o he Analy ic_pa ame e s
class. Fo example, he ange o he p ope y hasValue is e-
s ic ed o floa , when he class Bounce_ a e is i s domain;
he ange o he p ope y hasValue is es ic ed o da e, when
he class Da e_o _las _ isi is i s domain.
( ii) C ea e ins ances . Ins ances (indi iduals in OWL) co espond
o he specific da a ob ained om a specific e-shop. Indi id-
uals will be ob ained by mapping he da a om Google Ana-
ly ics, Piwik o compe i o s e-shops o RDF acco ding o he
on ology. Indi iduals can be also used in he on ology o de-
fine he exac membe s o a class. The ange o he p ope y
hasType is es ic ed o alues: ”ASIN, ”EAN o ”ISBN, when
i s domain is A icle_numbe . ASIN, EAN and ISBN a e hen
on ology indi iduals (see Table A.5 o u he explana ions).
4.1. On ology model
The p oposed on ology, called “wao.owl” (Web Analy ics On ol-
ogy), esul ing om he de elopmen p ocess desc ibed abo e has
a o al o 62 classes (g oups o indi iduals sha ing he same a -
ibu es), 61 objec p ope ies (bina y ela ionships be ween indi-
iduals), and 67 da a p ope ies (indi idual a ibu es), 33 es ic-
ion axioms and 3 indi iduals. The comple e on ology is a ailable
in WebP o égé eposi o y.
4
4 URL link h p://s an o d.io/1XHhHz
Fig. 2. Gene al o e iew o he WAO on ology. Con inuous a ows e e o sub class o . Do ed a ows e e o specific p ope ies.
Table 3
Analy ics_pa ame e s g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasB owse ∃ hasB owse .Thing  Analy ic_pa ame e s De ice
 
∀ hasB owse .B owse
hasCi y ∃ hasCi y.Thing  Analy ic_pa ame e s Loca ion Visi o
 
∀ hasCi y.Ci y
hasRegion ∃ hasRegion.Thing  Analy ic_pa ame e s Loca ion Visi o
 
∀ hasRegion.Region
hasCoun y ∃ hasCoun y.Thing  Analy ic_pa ame e s Loca ion Visi o
 
∀ hasCoun y.Coun y
hasCon inen ∃ hasCon inen .Thing  Analy ic_pa ame e s Loca ion
 
∀ hasCon inen .Con inen
hasSou ce ∃ hasSou ce.Thing  Analy ic_pa ame e s
 
∀ hasSou ce.Sou ce
Da a P ope ies Desc ip ion Logic
hasDa e ∃ hasDa e.Da a ypeLi e al  Analy ic_pa ame e s P ice P oduc _a ailabili y
 
∀ hasDa e.Da a ypeda eTimeS amp
hasHou ∃ hasHou .Da a ypeLi e al  Analy ic_pa ame e s
 
∀ hasHou .Da a ype ime
hasNe wo kDomain ∃ hasNe wo kDomain.Da a ypeLi e al  Analy ic_pa ame e s
 
∀ hasNe wo kDomain.Da a ypes ing
hasValue ∃ hasValue.Da a ypeLi e al  Analy ic_pa ame e s A icle_numbe P ice P oduc _a ailabili y
Fo simplici y, we desc ibe he e a ep esen a i e subse o main
classes including some o hei mos in e es ing objec and da a
p ope ies. These classes a e: Analy ics_pa ame e s, E-shop, Visi o ,
Page , and I em . Each class equi es a se o p ope ies o condi ions
in o de o be concep ualized. Tha is, an indi idual ha sa isfies
hose p ope ies is conside ed o be a membe o ha class.
-Analy ics_pa ame e s . Those a ibu es p o ided by Google
Analy ics and Piwik ha depend on ime. Each analy ic pa am-
e e has a alue ( hasValue in Table 3 ), which co esponds o
he da a p o ided by he analy ic ool, and a da e ( hasDa e ),
which co esponds o he da e when he da a was ob ained.
Subclasses in he on ology (  Analy ic_pa ame e s) a e, among
o he s: A e age_o de _ alue, A e age_pages_ isi ed_pe _session, A -
e age_session_du a ion, A e age_ ime_on_si e, Bounce_ a e, Con-
e sion_ a e, Numbe _o _ ansac ions, Numbe _o _landings, Num-
be _o _new_ isi o s, Numbe _o _page_ iews, Re enue_pe _ session
and To al_ e enue . Table 3 shows some ep esen a i e objec and
da a p ope ies o Analy ics_pa ame e s . Each analy ic pa ame-
e belongs o a da a ype. Fo ins ance, he alue o Num-
be _o _ ansac ions is an non-nega i e in ege and he alue o Con-
e sion_ a e is a floa . Da a ype es ic ions a e included in he on-
ology by means o da a p ope ies.
- E-shop. An e-shop has one o se e al pages and also
an e-shop’s owne . Each e-shop’s owne has an add ess.

Table 4
E-shop g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasVisi o >≡makesVisi >
−
∃ hasVisi o .Thing  E-shop
 
∀ hasVisi o .Visi o
hasNumbe O Visi o s ∃ hasNumbe O Visi o s.Thing  E-shop Page
 
∀ hasNumbe O Visi o s.Numbe _o _ isi o s
hasNumbe O Visi s ∃ hasNumbe O Visi s.Thing  E-shop Page Visi o
 
∀ hasNumbe O Visi s.Numbe _o _ isi s
isOwne O ∃ isOwne O .Thing  E-shop_owne
 
∀ isOwne O .E-shop
Da a p ope ies Desc ip ion logic
hasName ∃ hasName.Da a ypeLi e al  B owse Compe i o E-shop Goal I em
Ope a ing_sys em Page P oduc
 
∀ hasName.Da a ypes ing
hasURL ∃ hasURL.Da a ypeLi e al  Compe i o E-shop Page P ice
 
∀ hasURL.Da a ypes ing
Table 5
Visi o g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasDe ice ∃ hasDe ice.Thing  Visi o
 
∀ hasDe ice.De ice
hasNumbe O Visi s ∃ hasNumbe O Visi s.Thing  E-shop Page Visi o
 
∀ hasNumbe O Visi s.Numbe _o _ isi s
hasCi y ∃ hasCi y.Thing  Analy ic_pa ame e s Loca ion Visi o
 
∀ hasCi y.Ci y
makesVisi hasVisi o > ≡makesVisi >
−
∃ makesVisi .Thing  Visi o
 
∀ makesVisi .Visi
Da a p ope ies Desc ip ion logic
hasDaysSinceFi s Visi ∃ hasDaysSinceFi s Visi .Da a ypeLi e al  Visi o
 
∀ hasDaysSinceFi s Visi .Da a ypenega i eIn ege
hasDaysSinceLas O de ∃ hasDaysSinceLas O de .Da a ypeLi e al  Visi o
 
∀ hasDaysSinceLas O de .Da a ypenega i eIn ege
hasDaysSinceLas Visi ∃ hasDaysSinceLas Visi .Da a ypeLi e al  Visi o
 
∀ hasDaysSinceLas Visi .Da a ypenega i eIn ege
∃ IsRe u ningVisi o .Da a ypeLi e al  Visi o
 
∀ IsRe u ningVisi o .Da a ypeboolean
Table 6
Visi g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasNa iga ionS ep ∃ hasNa iga ionS ep.Thing  Visi
 
∀ hasNa iga ionS ep.Na iga ion_s ep
hasRe e e Keywo d ∃ hasRe e e Keywo d.Thing  Visi
 
∀ hasRe e e Keywo d.Re e e _keywo d
makesVisi hasVisi o > ≡makesVisi >
−
∃ makesVisi .Thing  Visi o
 
∀ makesVisi .Visi
Da a p ope ies Desc ip ion logic
hasDu a ion ∃ hasDu a ion.Da a ypeLi e al  Visi
 
∀ hasDu a ion.Da a ype ime
hasRe u ningVisi o ∃ hasRe u ningVisi o .Da a ypeLi e al  Visi
 
∀ hasRe u ningVisi o .Da a ypeboolean
A ibu es o he e-shop a e la i ude, longi ude and ime
zone. The e-shop’s owne can ha e compe i o s, who a e e-
shop’s owne s o o he e-shops. The analy ic pa ame e s o
an e-shop a e: a e age_o de _ alue, a e age_pages_ isi ed_pe _
session, a e age_ session_du a ion, a e age_ ime_on_si e, con e -
sion_ a e, da e_o _las _ ansac ion, numbe _o _cus ome s, numbe _
o _ ailed_ ansac ions, numbe _o _success ul_ ansac ions, numbe _o _
new_cus ome s, numbe _o _new_ isi o s, numbe _o _sessions_
by_medium, numbe _o _ ansac ions, numbe _o _unique_ isi o s,
numbe _o _uni s_sold, numbe _o _ isi o s, numbe _o _ isi s, pe -
cen age_o _new_sessions, e enue_pe _session, o al_ e enue , and
numbe _o _ e u ning_ isi o s . All he analy ic pa ame e s ela ed
o an e-shop a e ime dependen . The e o e, hey a e modeled
as classes and ela ed o he e-shop by he co esponding objec
p ope y. Table 4 shows a subse o p ope ies wi h classes in he
e-shop g oup as domain.
-Visi o and isi . Class isi o has wo subclases: cus ome
and New_ isi o . A cus ome is a isi o who makes a pu chase.
I i is he fi s pu chase o his cus ome , he/she is a new cus-
ome . Cus ome s ha e an add ess and name, whe eas isi o do
no . A isi o isi s he e-shop by using a de ice. The analy ic pa-
ame e s o isi o s a e: bounced_ a e, numbe _o _ isi s, and num-
be _o _ isi ed_pages . The analy ic pa ame e s o cus ome s a e
numbe _o _ ansac ions . Visi o s isi pages.
Visi s a e essen ial o cap u e he beha io o a isi o when
isi ing he e-shop. A isi has an en y page and an exi page. I
also has a e e e page, which is he way he isi o has accessed
he si e, i.e. sea ch engine, social ne wo k, web-ad e isemen e c.
I he e e e page is a sea ch engine, he keywo ds used o find
he si e a e also associa ed wi h he isi . A isi has a gi en du-
a ion. The a ibu es o a isi a e he imes when he en y page
and he end page we e accessed, du a ion, back link, whe he o
no an o de was placed and he o al goals con e ed du ing he
isi , numbe o ac ions, numbe o e en s and numbe o sea ches.
Du ing a isi , ansac ions a e made. A isi ollows a pa h which
has a nex page, a p e ious page and a numbe . The class Na i-
ga ion_s ep is used o model he pa h ha he use ollows om
he en y page o he exi page. Each na iga ion s ep has only one
Table 7
Page g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasNumbe O Visi s ∃ hasNumbe O Visi s.Thing  E-shop Page Visi o
 
∀ hasNumbe O Visi s.Numbe _o _ isi s
hasNumbe O Visi o s ∃ hasNumbe O Visi o s.Thing  E-shop Page
 
∀ hasNumbe O Visi o s.Numbe _o _ isi o s
hasTo alRe enue ∃ hasTo alRe enue.Thing  E-shop Page
 
∀ hasTo alRe enue.To al_ e enue
isOnPage ∃ isOnPage.Thing  I em
 
∀ isOnPage.Page
Da a p ope ies Desc ip ion logic
hasName ∃ hasName.Da a ypeLi e al  B owse Compe i o E-shop Goal I em
Ope a ing_sys em Page P oduc
 
∀ hasName.Da a ypes ing
hasURL ∃ hasURL.Da a ypeLi e al  Compe i o E-shop Page P ice
 
∀ hasURL.Da a ypes ing
hasTi le ∃ hasTi le.Da a ypeLi e al  Page
 
∀ hasTi le.Da a ypes ing
Table 8
I em g oup: objec and da a p ope ies.
Objec p ope ies Desc ip ion logic
hasI em ∃ hasI em.Thing  Page
 
∀ hasI em.I em
hasP ice ∃ hasP ice.Thing  I em Sha eP oduc Da a
 
∀ hasP ice.P ice
includes ∃ includes.Thing  O de
 
∀ includes.I em
isOnPage ∃ isOnPage.Thing  I em
 
∀ isOnPage.Page
Da a p ope ies Desc ip ion logic
hasCa ego y ∃ hasCa ego y.Da a ypeLi e al  I em
 
∀ hasCa ego y.Da a ypes ing
hasName ∃ hasName.Da a ypeLi e al  B owse Compe i o E-shop Goal I em
Ope a ing_sys em Page P oduc
 
∀ hasName.Da a ypes ing
hasI emID ∃ hasI emID.Da a ypeLi e al  I em
 
∀ hasI emID.Da a ypenonNega i eIn ege
hasQuan i y ∃ hasQuan i y.Da a ypeLi e al  I em
 
∀ hasQuan i y.Da a ypenonNega i eIn ege
a ibu e numbe . Tables 5 and 6 show he p ope ies wi h classes
in he isi o and isi g oup as domain, espec i ely.
-Page. Pages con ain i ems, i.e. p oduc and/o se ices o be
sold. The analy ic pa ame e s o Page a e: a e age_o de _ alue,
a e age_ ime_on_page, bounce_ a e, da e_ o _las _ isi , num-
be _o _exi s, numbe _o _landings, numbe _o _new_ isi o s, numbe _
o _page_ iews, numbe _o _ e u ning_ isi o s, numbe _o _sessions_
by_medium (mediums a e di ec link, social media and sea ch
engine), numbe _o _sessions, numbe _o _unique_page_ iews, num-
be _o _unique_ isi o s, numbe _o _uni s_sold, numbe _o _ isi o s,
numbe _o _ isi s, e enue_ pe _session_and_ o al_ e enue . A ibu es
o page a e i le and URL. A se ies o ep esen a i e p ope ies
whose domain is page a e shown in Table 7 . In e es ingly, we can
obse e in his able ha he p ope y hasTo alRe enue is ela ed
o he Page , as well as he o he whole e-shop, as his alue can
be calcula ed o bo h classes.
-I em. As commen ed be o e, an I em is a p oduc o a se ice
which is sold in an e-shop . Specific i ems o an e-shop a e modeled
by defining a domain on ology o a specific domain, i.e., a el,
books, music, e c. Table 8 con ains some ep esen a i e objec and
da a p ope ies o class i em . Acco ding o his, I ems ha e a p ice
( hasP ice ). P ices a e alid on a ce ain da e. The e o e, a ibu es
o p ices a e alue, cu ency and he da e o he p ice alidi y .
The a ibu es o I ems a e ca ego y and whe he o no i has been
dele ed. P oduc s ha e a manu ac u e . The a ibu es o p oduc s
a e: name, ype, a ailabili y on a specific da e and a icle numbe .
The a icle numbe can be “ASIN”, “EAN” o “ISBN”.
4.2. Da a sou ces: mapping and que ying
As we explained in Sec ion 3 , we ha e ocused on h ee main
sou ces o da a coming om di e en web acking me hods,
namely: Google Analy ics, Piwik, and specific web sc apping me h-
ods in he scope o SME E-Compass p ojec .
The p ocess o ansla ing he collec ed da a om di e en
sou ces o RFD is ca ied ou by means o mapping unc ions. Each
da a sou ce has a di e en se o me hods o ga he , ha monize,
s o e and p o ide access o he analy ical da a. The e o e, a di -
e en se o mapping unc ions is equi ed o pa se he in o ma-
ion coming om each da a sou ce o RDF, acco ding o he on-
ology. Fig. 3 illus a es an gene al o e iew o he mapping p o-
cess o s o e da a om di e en sou ces in a common RDF eposi-
o y. Each se o mappings is hen composed by unc ions o ans-
la e he a ibu es wi h hei alues in o hei co esponding iple
o m in RDF. In ac , o mos o he a ibu es, a co esponding
mapping unc ion has been de eloped, in ol ing i s co espond-
ing class in he on ology. Ne e heless, as a numbe o analy ic
a ibu es sha es a common s uc u e in he on ology, hey ha e
been mapped by using gene ic unc ions, hence aking ad an age
o he on ology’s design.
Fig. 3. Gene al o e iew o he mapping p ocess injec ing da a om di e en sou ces in o he RDF eposi o y.
4.2.1. Google Analy ics
Google Analy ics
5 is a pa ially ee web analy ics se ice ha
p o ides s a is ics and basic analy ical ools o Sea ch Engine Op-
imiza ion (SEO) and ma ke ing pu poses. The se ice is a ailable
o anyone wi h a Google accoun , al hough ad anced e-comme ce
unc ionali ies a e only a ailable o es ic ed use s. Google Ana-
ly ics is gea ed owa d small and medium-sized e ail websi es.
The web acking p ocedu e in Google Analy ics is pe o med
by a “snippe ” o digi al oo p in componen , ha p o ides he
de elope wi h an API o unc ions o accessing o each a ibu e
alue. This digi al oo p in is a small piece o Ja aSc ip code ha
is pas ed in o he e-shops HTML sou ce code and deployed in he
web se e whe e he e-shop is hos ed. I ac i a es Google Ana-
ly ics acking by inse ing he Ja aSc ip ga.js / analy ics.js
in o he page. As illus a ed in Fig. 3 , he Ja aSc ip componen is
hen ins an ia ed by ou mapping unc ions by means o a se ies
o ja a classes o gene a e RDF iples.
Table A.1 in Appendix A con ains he se o Google Analy ics
a ibu es ha a e cu en ly acked by ou seman ic app oach. In
his able, each a ibu e is lis ed wi h ega ds o i s co esponding
on ology class, da a ype, and desc ip ion. This is a ep esen a i e
subse o he whole se o possible a ibu es (and i s combina-
ions) in he Google Analy ics’s API specifica ion, ha co e s all ou
p elimina y equi emen s o isi o ’s beha io and p oduc s’ anal-
ysis. Howe e , i is wo h men ioning ha he p oposed on ology
can be easily ex ended o conside any o he a ibu es wo ked
wi h Google Analy ics.
4.2.2. Piwik
Piwik
6
is a ee and open sou ce web analy ics applica ion un-
ning on a PHP/MySQL web se e . Piwik acks online isi s o one
o mo e websi es and displays epo s on hese isi s o analysis.
Piwik analy ic ea u es include, among o he s: eal- ime da a up-
da es, ee ad anced e-comme ce analy ics ea u es, goal con e -
5 h p://www.google.com/analy ics/
6 h p://piwik.o g/
sion acking, e en acking, geoloca ion, pages ansi ions iews
and page o e lay.
Simila ly o Google Analy ics, he web acking p ocedu e in
Piwik is also pe o med by a digi al oo p in sc ip , ha is al-
loca ed in he e-shops HTML sou ce code. In he case o Piwik,
he analy ical da a is au oma ically s o ed in a ela ional da abase
(SQL). The e o e, as we ha e he possibili y o access o his ela-
ional da abase, we ha e de eloped he mapping unc ions o di-
ec ly que y he analy ic a ibu es. These a ibu es a e desc ibed
in Tables A .2–A .4 o Appendix A wi h ega ds o hei co espond-
ing on ology classes. The ob ained da a is hen ansla ed o RDF
acco ding o he on ology by means o specific mapping me hods,
as shown in Fig. 3 .
4.2.3. Web sc apping me hods
In he scope o he SME E-Compass p ojec , he e exis a se ies
o me hods o sc aping p oduc and p ice da a om he compe i-
o s e-shop websi es. This way, a gi en e-shop’s owne is able o
compa e hei p oduc s’ p ices wi h hose ones o hei di ec com-
pe i o s au oma ically.
This specific unc ionali y p o ides a REST API se ice om wi h
we can ob ain a ibu es o compe i o ’s p ofile in JSON
7 o ma ,
which is a compac and easily eadable da a o ma o he pu pose
o da a exchange. Table A.5 con ains he compe i o s a ibu es ha
a e mapped o RDF in ou seman ic model (see Fig. 3 ), wi h e-
ga ds o he co esponding on ology classes.
4.2.4. RDF eposi o y
Finally, an RDF eposi o y is used o in eg a e he analy ic da a
collec ed and mapped om he di e en sou ces. The e o e, by
means o an SPARQL endpoin , i is now possible o que y he an-
aly ic da a unambiguously and independen ly o he sou ce.
As an ins ance o da a access, le us conside an scena io in
which he analy ic module equi es in o ma ion conce ning he
7 h p://json.o g
Fig. 4. Example o SPARQL que y ha e u ns disagg ega ed da a a ibu es, as he ones p o ided by Piwik, as well as calcula ed me ics, as hose ob ained om Google
Analy ics.
Table 9
Two samples o he que y esul ( Fig. 4 ) o a ce ain ime slo
(day 2015-10-23) o a eal e-shop.
A ibu e/me ic Visi 75688 Visi 75692
imes amp 14 :19:44 14 :21:41
isi _ o al_sea ches 0 0
isi _ o al_e en s 0 0
isi _ o al_du a ion 2071 12
isi _ o al_goal_con e ed 1 0
o al_bounce_ a e 52.6066
o al_con e sion_ a e 34.1232
o al_numbe _o _en ies 211
o al_numbe _o _new_ isi o s 145
isi s o a gi en e-shop, in a ce ain da e o pe iod o ime. The e-
qui ed in o ma ion o isi s should consis o bo h: disagg ega ed
da a a ibu es, as he ones p o ided by Piwik, and calcula ed me -
ics, as hose ob ained om Google Analy ics.
The SPARQL que y ep esen ed in Fig. 4 unifies he encoding o
such logic, o which a couple o esul samples a e displayed in
Table 9 . In conc e e, hese esul s co espond o wo consecu i e
isi s o he e-shop wi h ID < eshop-id > , ha we e pe o med
a da e 2015-10-23. The isi IDs a e 75688 and 75692, and hey
we e cap u ed a imes amps 14:19:44 and 14:21:41, espec i ely.
As shown in his able, he isi wi h a p olonged du a ion led o
one goal con e sion (usually a success ul sale), whe eas he isi
wi h a sho du a ion finished wi hou any con e sion, which ep-
esen s a isi o ha lea es he si e p ema u ely.
In he case o agg ega i e a ibu es, hey a e calcula ed o
all he isi s in he ime pe iod o he SPARQL que y. The e o e,
as shown in he second hal o Table 9 , he e-shop egis e ed a
bounce a e close o 53% wi h con e sion a e
8 o 34.12%, ha co -
esponds o all isi s, bounces and pu chases o he que ies ime
pe iod.
Ano he impo an a ibu e is he numbe o new isi o s, ha
o his e-shop and o his da e is 145, e. g., 68.72% o o al en ies.
This in o ma ion could be now used o eed p edic i e algo i hms
ha help he e-me chan o adop a gi en ma ke ing s a egy o
ca ch clien s.
In o de o au oma ize and simpli y he accesses o he s o ed
da a, ou seman ic app oach includes a specific REST API se ice
wi h me hods ha implemen p edefined SPARQL que ies. These
me hods a e used as inpu o he da a mining algo i hms as de-
sc ibed in he ollowing s ep o alida ion.
As an addi ional ad an age o his seman ic app oach, i is pos-
sible o connec ou RDF eposi o y wi h o he /s ex e nal open
linked da a eposi o y/ies. In his ega d, a minimum adap a ion
has o be done in e ms o deciding which class/classes a e di ec ly
linked om he wo eposi o ies wi h simila seman ic meaning. In
ac , his is one o he mos powe ul ea u es when using he se-
man ic s uc u e induced by he on ology.
8 Con e sion a e: p opo ion o isi o s con e ed in o paying cus ome s.