182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 178
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
Ma io Sc occa, Alessio Ca enini, Valen ina Ca ie o,
and I ene Celino
Ce iel –I aly
HAIB idge 2025
1s Wo kshop on B idging Hyb id (A i icial)
In elligence and he Seman ic Web, co-loca ed wi h
ISWC 2025, Na a, Japan
Adap i e Fo ms o Semi-au oma ic
Knowledge Collec ion in RDF
apid- iples
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
P oblem Add essed
‣To educe inaccu acies o a lack o con ex , AI
applica ions hea ily bene i om s uc u ed
knowledge
‣Collec ing he knowledge and building a
Knowledge G aph acco ding o e e ence
on ologies is a challenge equi ing domain
expe s' in ol emen
‣We p opose he apid- iples ool o suppo
expe -guided knowledge c ea ion and AI-
enabled knowledge comple ion om exis ing
uns uc u ed documen s
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
Wo k lows o Knowledge Collec ion
•W1 Taci Knowledge Collec ion: ele an knowledge in he
minds o domain expe s and no ye documen ed. The use
mus be guided in a icula ing and o malising i acco ding o
he a ge on ology.
•W2 Knowledge Comple ion om Uns uc u ed Sou ces:
ini ial knowledge is au oma ically ex ac ed om
uns uc u ed con en using AI-based ools. A human-in- he-
loop (HITL) p ocess ensu es ha he esul ing knowledge is
accu a e, comple e and seman ically consis en wi h he
a ge on ology.
•W3 Enhance Au oma ic Knowledge Ex ac ion Sys ems:
s uc u ed knowledge, alida ed and e iewed by use s, is
used as aining o con ex ual da a o enhance he accu acy
o au oma ic knowledge ex ac ion solu ions.
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
Rula, Anisa, e al. "Anno a ion and Ex ac ion o Indus ial P ocedu al Knowledge om
Tex ual Documen s." P oceedings o he 12 h Knowledge Cap u e Con e ence 2023. 2023.
Au oma ic Knowledge Collec ion in RDF
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
Manual Knowledge Collec ion in RDF
‣Au oma ic anno a ion om documen s ails
o unde s and he knowledge implici ly
de ined by he documen s uc u e (e.g.,
e e ence o sub-p ocedu es in o he
sec ions)
‣Au oma ic anno a ion should be ine- uned
o speci ic documen s uc u es, bu o en
documen s ollow he e ogeneous empla es
‣On oPawls (h ps://gi hub.com/ce iel/on o-
pawls) o e s a ool o anno a e PDFs
acco ding o a gi en on ology
‣Manual e o equi ed by use s is high
Rula, Anisa, e al. "Anno a ion and Ex ac ion o
Indus ial P ocedu al Knowledge om Tex ual
Documen s." P oceedings o he 12 h
Knowledge Cap u e Con e ence 2023. 2023.
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
Semi-au oma ic Knowledge Collec ion in RDF
•W1 Taci Knowledge
Collec ion
•W2 Knowledge
Comple ion om
Uns uc u ed Sou ces
•W3 Enhance Au oma ic
Knowledge Ex ac ion
Sys ems
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
Semi-au oma ic Knowledge Collec ion in RDF
•W1 Taci Knowledge
Collec ion
•W2 Knowledge
Comple ion om
Uns uc u ed Sou ces
•W3 Enhance Au oma ic
Knowledge Ex ac ion
Sys ems
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 178
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
8
Why apid- iples?
‣To guide he use s in unde s anding which
in o ma ion is necessa y (e.g., aci
knowledge) and how o model i
‣To hide he complexi y o he RDF
ep esen a ion om he use
‣To enable usage o he ool seamlessly wi h
o he componen s and suppo he
p oposed wo k lows o semi-au oma ic
knowledge collec ion
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 278
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
‣The objec i e is o enable decoupled de elopmen
and seamless in eg a ion o di e en componen s
o en no able o deal di ec ly wi h he RDF
ep esen a ion
‣JSON-based o ma in oduced elying on he
seman ics o he a ge on ology. P o ides a common
“ aming” o e he RDF g aph ep esen a ion.
‣A JSON Schema de ines alid documen s and enables
au oma ic alida ion
‣A single KG cons uc ion p ocess can be implemen ed
elying on he JSON Schema ( o simple cases a JSON-
LD con ex is su icien )
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
In e media e Exchange Fo ma
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 178
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
apid- iples P elimina y E alua ion
R A P I D - T R I P L E S ( H A I B R I D G E ’ 2 5 )
We un a p elimina y quali a i e e alua ion o he apid- iples
ool wi h Beko use s:
‣The use s we e able o gene a e a alid RDF ep esen a ion o
LOTO p ocedu es using PKO o di e en machines in he
ac o y
‣The domain expe s app ecia ed he manual collec ion h ough
he adap i e o m, as i e ec i ely guided hem in documen ing
addi ional aci knowledge while p o iding a be e use
expe ience wi h espec o pape o abula -based app oaches
‣Sa e y manage s highligh ed he highe quali y o he
gene a ed p ocedu es, and LOTO ope a o s alued he
possibili y o ha ing access o mo e de ails du ing he
p ocedu e execu ion.
www.pe ks-p ojec .eu
www.pe ks-p ojec .eu
THANK YOU!
This p ojec has ecei ed unding om he Eu opean Union’s Ho izon Eu ope esea ch
and inno a ion p og amme unde g an ag eemen No 210906180
182 211 131
103 183 0
47 60 182
61 177 166
127 211 203
183 233 229
235 249 248
46 133 124
4 77 92
103 183 0
248 203 14
#1
#2
#3
#4
#5
8 134 53
103 183 0
150 189 71
182 211 131
210 228 178
#1
#2
#3
#4
#5
P ima y
5 3 14 2
5 3 1 4 2
GRAPHS
Sequence o use
PERKS IDENTITY
46 133 124
224 83 72
00 00 00
Tex
Ti le 1
Ti le 2
TEXT
Seconda y
Sequence o use
BACKGROUND
RAG
46 133 124
4 77 92
8 134 53
Ma io Sc occa
ma io.sc occa@ce iel.com
apid- iples
•Adap i e o m-based in e ace cus omizable ia JSON Schema
•Decoupling JSON ou pu and li ing p ocess o RDF suppo s he
implemen a ion o semi-au oma ic knowledge collec ion wo k lows
•Use case conside ing p ocedu al knowledge collec ion
•Clien implemen a ion a ailable on Gi Hub
Fu u e wo k: In es iga e mo e complex wo k lows; pe o m b oade
and quan i a i e use e alua ion; in eg a e decla a i e mapping ules
and gene a i e AI o u he educe manual in e en ion
pko- apid- iples
•Check he demo wi h PKO
•Access he apid- iples
empla e eposi o y
•Check how o cus omize
he empla e
