knowledge graphs and enterprise...
TRANSCRIPT
![Page 1: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/1.jpg)
Georg GottlobUniv. of Oxford, TU Wien
& DeepReason.ai
Knowledge Graphs and Enterprise AIThe Promise of an Enabling Technology
![Page 2: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/2.jpg)
VADAEPSRC
PROJECT
Manchester Oxford
Edinburgh
Georg GottlobUniv. of Oxford, TU Wien
& DeepReason.ai
Knowledge Graphs and Enterprise AIThe Promise of an Enabling Technology
![Page 3: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/3.jpg)
Knowledge Graphs as Large “World” KBs
Cyc [Lenat &Guha 1989]: “comprehensive ontology and knowledge base of
everyday common sense knowledge”.
Freebase [Bollacker et al. 2007] : “online collection of structured data harvested from many sources, including user-submitted wiki contributions”.
Google Knowledge Graph [Singhal 2012] + K.Vault [Dong et al. 2014]: “KB used by Google to enhance its search engine's search
results with semantic-search information gathered from a wide variety of sources”.
DBpedia [Auer et al. 2007]. Yago [Suchanek et al 2007]both generate structured ontologies from Wikipedia.
Wikidata [Vrandečić 2012, Krötzsch+V. 2014] open knowledge basethat can be read and edited by both humans and machines.
![Page 4: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/4.jpg)
More Specialized Knowledge Graphs
Facebook Knowledge Graph: Social graph with people, places and things + information from Wikipedia
Amazon Knowledge Graph: Started as product categorization ontology
Wolfram KB: World facts + mathematics
Factual: Businesses & places
Megagon (Recruit Inst.): People, skills, recruiting
Central Banks: Company register – ownership graph
Credit Rating Agencies …
Thousands of medium to large size companies now want their own corporate knowledge graph. This not just for semantic indexing and search, but for advanced reasoning tasks on top of machine learning.
![Page 5: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/5.jpg)
Reasoning in Knowledge Graphs
Ontology / Rules
EDB/ABox
EDB+IDB
Reasoning tasks are required that cannot be expressed by description logics, and cannot be reasonably managed by relational DBMS, nor by graph DBMS.
Many still think that DLs or graph databases suffice. However:
![Page 6: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/6.jpg)
Wikidata contains the statement :
Taylor was married to Burton starting from 1964 and ending 1974
Example: Wikidata Marriage Intervals
This can be represented in relational DB or Datalog-notation by :
[Krötzsch DL 2017]
∀ u,v,x,y. married(u,v,x,y) → married(v,u,x,y)
Symmetry rule for marriage intervals in Datalog:
This cannot be expressed in DLs!
married(taylor,burton,1964,1974)
Note: In what follows, we will often omit universal quantifiers.
![Page 7: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/7.jpg)
0.138
0.093
0.003 0.083
0.007
0.51
0.24
0.25
0.381
0.006
0.003
0.073
0.093
0.32
0.17
0.22
0.472
0.7
0.081
0.138
0.007
0.0070.093 0.013
0.191
0.65
1
0.38
0.128
0.0820.195
0.007
0.5920.932
0.334
0.875
0.131
0.532
0.982
0.4430.1120.452
0.143 0.013
0.048
0.393
0.878
Example: Controlling Companies
![Page 8: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/8.jpg)
Example: Controlling Companies
company(x) → own(x,x).
own(x,y,w), w>0.5 → control(x,y).
control(x,y), own(y,z,w), v=msum(w,y), v>0.5 → control(x,z).
This cannot be expressed in DLs and only clumsily in SQL and Graph DBMS!
x controls y if x directly holds over 50% of y, or x controls a set of companies that jointly hold over 50% of y
![Page 9: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/9.jpg)
Example: My Creditworthiness
![Page 10: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/10.jpg)
Example: My Creditworthiness
up to £10,000 £8,500 £12,000
up to EUR 10,000 up to EUR 20,000
£ 8,000 £ 12,500 EUR 14,000
£500
![Page 11: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/11.jpg)
People who live in a joint household with someone who does not pay their bills are likely to fail repaying their own debts.
A machine-learning program has “reasonably” learned:
This ethically questionable rule was applied to wrong data.
Explanation
![Page 12: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/12.jpg)
People who live in a joint household with someone who does not pay their bills are likely to fail repaying their own debts.
A machine-learning program has “reasonably” learned:
This ethically questionable rule was applied to wrong data.
A human credit rating expert would instead use of the rule:
Explanation
If property owners move into their recently bought one-familyproperty, then the previous occupiers have most likely moved out.
(Such updates are often missing in the database)
This rule can be used to update the database before applying machine learning.
![Page 13: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/13.jpg)
Knowledge Graph Management Systems (KGMS)
KGMS combine the power of rule-based reasoning with machine learning over Big Data:
KGMS = KBMS + Big Data + Analytics
Misusing the lateralization thesis for illustration
symbolic subsymbolictaught self-learned
![Page 14: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/14.jpg)
symbolic subsymbolictaught self-learned
Fast thinking
Slow thinking
Grandma: “Fly agarics are poisonous mushrooms. If you eat a poisonous mushroom, you may die”.
Yikes, a fly agaric!
![Page 15: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/15.jpg)
Desiderata for KGMS According to our Philosophy
No extra permanent data repository or database/DBMS- Uses (possible multiple) existing company data repositories/databases- Can query and update these – streaming into main memory for reasoning - No data migration necessary
Multiple data models possible. - Relational, graph, RDF, …- Reasoning engine interprets all data relationally (by Datalog facts)
High expressive power of reasoning language; express at least:- Full Datalog with full recursion and stratified negation- Graph navigation- Aggregate functions- Description logics such as: DL-Lite (OWL 2 QL), EL, F-Logic Lite- SPARQL under RDFS or OWL2 QL Entailment Regimes
Good complexity and scalability - Tractability guarantee for main formalism- Highly efficient, and highly parallelizable language fragments
Support for machine learning, analytics, and collaborative filtering- APIs to standard ML and analytics packages (do not reinvent the wheel)- Provide system support for graph analysis (e.g. balanced separators), and typical
functions such as argmin (with grad. desc.), eigenvector, pagerank, simrank, etc.
![Page 16: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/16.jpg)
Knowledge Graph Management Systems
a diverse new field – many systems with different capabilities
Graph Path
![Page 17: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/17.jpg)
Graph database supporting SPARQL and Prolog reasoning
Apache Cassandra-based KGMS providing schema support based on the Entity Relationship model
Knowledge Graph-as-a-Service
Data source-agnostic KGMS supporting ontological and recursive reasoning based on Datalog
Leading graph database system
RDF-based unifying data-integration platform
SPARQL 1.1-graph database-based end-user-oriented platform
Azure-based computation-focused platform
RDF and OWL-based metadata management solution.
Analysis along many dimensions possible
![Page 18: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/18.jpg)
Migration necessary?
Graph database supporting SPARQL and Prolog reasoning
Apache Cassandra-based KGMS providing schema support based on the Entity Relationship model
Knowledge Graph-as-a-Service
Data source-agnostic KGMS supporting ontological and recursive reasoning based on Datalog
Leading graph database system
RDF-based unifying data-integration platform
SPARQL 1.1-graph database-based end-user-oriented platform
Azure-based computation-focused platform
RDF and OWL-based metadata management solution.
![Page 19: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/19.jpg)
Graph database supporting SPARQL and Prolog reasoning
Apache Cassandra-based KGMS providing schema support based on the Entity Relationship model
Knowledge Graph-as-a-Service
Data source-agnostic KGMS supporting ontological and recursive reasoning based on Datalog
Leading graph database system
RDF-based unifying data-integration platform
SPARQL 1.1-graph database-based end-user-oriented platform
Azure-based computation-focused platform
RDF and OWL-based metadata management solution.
Principle Data Format / Backend
![Page 20: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/20.jpg)
Graph database supporting SPARQL and Prolog reasoning
Apache Cassandra-based KGMS providing schema support based on the Entity Relationship model
Knowledge Graph-as-a-Service
Data source-agnostic KGMS supporting ontological and recursive reasoning based on Datalog
Leading graph database system
RDF-based unifying data-integration platform
SPARQL 1.1-graph database-based end-user-oriented platform
Azure-based computation-focused platform
RDF and OWL-based metadata management solution.
Analysis along many dimensions possible
![Page 21: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/21.jpg)
Vadalog KGMS Being Built at Oxford
• VADA = Value-Added DAta
• General architecture of VADALOG system
• Core reasoning language VADALOG = Warded Datalog + extensions
• Connectivity: Some plug-ins
Current Team Members
Tim Furche
![Page 22: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/22.jpg)
+ special features:argmin, sampling, graph libraries (e.g.separators),
matrix ev, simrank, pagerank, …
VADALOG CORE REASONING
WEB
RuleRepository
![Page 23: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/23.jpg)
Vadalog: The Core Reasoning Language
Core Vadalog = full Datalog + restricted use of + stratif. negation +
• Data exchange, data integration• Data extraction • Reasoning with RDF Wikidata example • Ontology querying (DL-Lite, EL, etc.)• Data anonymization• Duplicate handling• Automated product configuration• Conceptual Modeling (e.g., UML)
Why existential quantifiers in rule heads?
![Page 24: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/24.jpg)
person(FirstName, Lastname, Birthdate)
employee(Lastname, Firstname, Address)
employee(X,Y,Z) → W person(Y,X,W)
Data Exchange, Data Provisioning, Data Wrangling
Source Schema S Target Schema T
I ?
st t
[Fagin, Kolaitis, Miller & Popa, 2003]; [Arenas et al., 2014]
![Page 25: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/25.jpg)
Object Creatione.g. in web data extraction
PRODUCT
Toshiba_Protege_cx
Dell_25416
Dell_23233
Acer_78987
PRICE
480
360
470
390
![Page 26: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/26.jpg)
PRODUCT
Toshiba_Protege_cx
Dell_25416
Dell_23233
Acer_78987
PRICE
480
360
470
390
T1 T2
Object Creatione.g. in web data extraction
![Page 27: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/27.jpg)
PRODUCT
Toshiba_Protege_cx
Dell_25416
Dell_23233
Acer_78987
PRICE
480
360
470
390
T1 T2
Object Creatione.g. in web data extraction
![Page 28: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/28.jpg)
PRODUCT
Toshiba_Protege_cx
Dell_25416
Dell_23233
Acer_78987
PRICE
480
360
470
390
table(T1),
table(T2),
sameColor(T1,T2),
isNeighbourRight(T1,T2)
T tablebox(T),
contains(T,T1),
contains(T,T2).
T1 T2
Object Creatione.g. in web data extraction
![Page 29: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/29.jpg)
In the RDF-like “graph” notation this tuple is broken up into several triples (here represented as logical facts):
spouse1(u,y1) ∧ spouse2(u,y2) ∧ start(u,y3) ∧ end(u,y4) →
∃v.spouse(v,y1) ∧ spouse1(v,y2) ∧ start(v,y3) ∧ end(v,y4)
This symmetry rule for marriage intervals now becomes:
spouse1(k1,taylor),
spouse2(k1,burton),
start(k1,1964),
end(k1,1974)
spouse1(k2,burton),
spouse2(k2,taylor),
start(k2,1964),
end(k2,1974)
∀ u,v,x,y. married(u,v,x,y) → married(v,u,x,y)
Reasoning with RDF – Foreign Key Creation
![Page 30: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/30.jpg)
The DL-Lite Family
Popular family of DLs with low (AC0) data complexity
[Calvanese, De Giacomo, Lembo, Lenzerini & Rosati, J. Autom. Reasoning 2007]
Description Logics & Ontological Reasoning
DL-Lite TBox First-Order Representation (Datalog§)
DL-Litecore
professor v 9teachesTo
professor v :student
DL-LiteR (OWL 2 QL)
hasTutor¡ v teachesTo
DL-LiteF
funct(hasTutor)
8X professor(X) 9Y teachesTo(X,Y)
8X professor(X) student(X) ?
8X8Y hasTutor(X,Y) teachesTo(Y,X)
8X8Y8Z hasTutor(X,Y) hasTutor(X,Z) Y = Z
![Page 31: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/31.jpg)
Datalog[]: Full Datalog augmented with -quantifier
[Beeri & Vardi, 1981]; [J. Mitchell 1983] [Chandra & Vardi 1985];
[Calì, G., & Kifer, 2008]; [Baget, Leclère & Mugnier, 2010]
Unfortunately:
Theorem: Reasoning (𝐾𝐵 ⊨ 𝑞) with Datalog[] is undecidable.
Finding expressive decidable/tractable fragments has become a topic
of intensive research over the last 10 years.
Datalog : Datalog[,⊥,strat, …] subject to syntactic restrictions.
Vadalog: member of the Datalog family admitting efficientreasoning methods.
![Page 32: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/32.jpg)
Guarded
Sticky
DL-LiteR
Weakly-Sticky
Weakly-(frontier-) guarded
Main Decidable Datalog± Languages
Linearℰℒℋ𝐼
BTS
FUS
![Page 33: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/33.jpg)
Guarded
Sticky
DL-LiteR
Weakly-Sticky
Weakly-(frontier-) guarded
Main Decidable Datalog± Languages
Linearℰℒℋ𝐼
BTS
FUS
![Page 34: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/34.jpg)
Datalog[,…]
weakly frontier-guarded Datalog[,⊥,strat]
Core Vadalog
= warded Datalog[,⊥,strat]
SPARQL + OWL 2QL
UNDECIDABLE
EXPTIME
PTIME
Data complexity
Datalog[⊥,strat]
Linear Datalog[,⊥]
AC0
![Page 35: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/35.jpg)
Datalog[,…]
weakly frontier-guarded Datalog[,⊥,strat]
Core Vadalog
= warded Datalog[,⊥,strat]
SPARQL + OWL 2QL
UNDECIDABLE
EXPTIME
PTIME
Data complexity
Datalog[⊥,strat]
Linear Datalog[,⊥]
AC0
![Page 36: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/36.jpg)
T[3], P[1], Q[2]
T[2], P[2], Q[1]
Affected Positions
Vadalog is based on Warded Rules
A Datalog program is warded if for each rule body:
• all dangerous variables jointly occur in a single „ward“ atom, and
• this ward shares only unaffected variables with the other body-atoms
Core Vadalog = warded Datalog[,⊥,strat]
P(X,Y), S(Y,Z) ∃W T(Y,X,W)
P(X,Y) ∃Z Q(X,Z)
T(X,Y,Z) ∃W P(W,Z)
![Page 37: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/37.jpg)
Examples of Warded Datalog Rules
spouse1(x,y1) ∧ spouse2(x,y2) ∧start(x,y3) ∧ end(x,y4) →
∃v. spouse2(v,y1) ∧ spouse1(v,y2) ∧
start(v,y3) ∧ end(v,y4)
1. Symmetry rule for marriage intervals:
2. : OWL 2 QL description logic
DL-Lite TBox Representation in Vadalog
DL-Litecore
professor v 9teachesTo
professor v :student
DL-LiteR (OWL 2 QL)
hasTutor¡ v teachesTo
8X professor(X) 9Y teachesTo(X,Y)
8X professor(X) student(X) ?
8X8Y hasTutor(X,Y) teachesTo(Y,X)
![Page 38: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/38.jpg)
- Datalog with full recursion and stratified negation
- Description logics: DL-Lite Family, in particular, OWL 2 QL, EL, F-Logic Lite
- SPARQL under RDFS and OWL 2 QL Entailment Regimes
Theorem
Vadalog can express:
![Page 39: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/39.jpg)
- Datalog with full recursion and stratified negation
- Description logics: DL-Lite Family, in particular, OWL 2 QL, EL, F-Logic Lite
- SPARQL under RDFS and OWL 2 QL Entailment Regimes
Theorem
Vadalog can express:
Moreover: All queries of iBench can be expressed in Vadalog!
![Page 40: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/40.jpg)
Further Language Features (selection)
Data types and associated operations & expressions: integer, float, string, Boolean, date, sets.
Monotonic aggregations: min, max, sum, prod, countwork even in presence of recursion while preserving monotonicity of set-containment
Example: Company Control
own(x,y,w), w>0.5 → control(x,y);
control(x,y),own(y,z,w),
v=msum(w,y), v>0.5 → control(x,z).
Probabilistic reasoning: facts and rules can be adorned withweights. Marginal weights for derived facts will becomputed assuming independence.
Equality (EGDs, functional dependencies) if non-conflicting.
![Page 41: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/41.jpg)
Rules can be uncertain
@weight(0.6) company(C) → ∃C1 own(C,C1).
@weight(0.5) own(C,S), holding(C) → subsidiary(S).
A Soft Vadalog rule has a weight
Similar to Markov Logic Network, but Soft Vadalogis not full First Order Logicallows recursive definitionshas unrestricted domain
![Page 42: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/42.jpg)
Database Interface
@bind("Own", "rdbms", "companies.ownerships").
@qbind("Own", "graphDB", "MATCH (a)-[o:Owns]->(b) RETURN a,b,o.weight").
Cypher query (Neo4j)
@bind("q","data source", "schema","table").
@update("q",{1,3,4,5}).
![Page 43: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/43.jpg)
Machine Learning, Big Data Analytics, NLP & Data Visualization
We are currently experimenting with different tools and different types of interfaces and interactions.
Interaction Model 1
Machine Learning
i
o
Interaction Model 2
Machine Learning
i
P(...)
Interaction Model 3
Machine Learning
i
LEARNED RULES
![Page 44: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/44.jpg)
@qbind("Own", "oxpath",
"doc(‘http://company_register.com/ownerships’)
/descendant::field()[1]/{$1}
/following::a[.#=’Search’]/{click/}
/(//a[.#=’Next’]/ {click/})∗//div[@class=’c’]: [./span[1]:][./span[3]:]")
Web Data Extraction& IoT
Interfacing KG to OXPath;Binding OXPath to Datalog
[Furche, T., Gottlob, G., Grasso, G., Schallhart, C., & Sellers, A. (2013). OXPath: A language for scalable data extraction, automation, and crawling on the deep web. The VLDB Journal, 22(1), 47-72. ".]
![Page 45: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/45.jpg)
Core Algorithms
• Bottom-up chase processing with „aggressive“ termination strategy
• Top-down query processing (currently under implementation)
• Advanced program rewriting and optimization techniques
• Efficient & highly scalable cache managmt., query plan optimization
• Recent evaluation shows the system is extremely competitive
For more details see Luigi Bellomarini, Emanuel Sallinger, Georg Gottlob: The Vadalog System: Datalog-based Reasoning for Knowledge Graphs. PVLDB 11(9) 2018
![Page 46: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/46.jpg)
Parser
Query
manager
Logical optimizer
Planner
Execution plan optimizer
Record manager
Cache manager Termination manager In-memory indexer
Interfaces (REST, JDBC, API, GUI, …)
Expressions
evaluatorAggregator
Probabilistic
reasoning
Warded Datalog±
Nearly-linear
Datalog±
PVLDB 11(9) 2018.
![Page 47: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/47.jpg)
Input Scan
Join Scan
Linear Datalog± Scan
Output Scan
In-Memory Stream Processing
Similar to Volcano iterator model
![Page 48: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/48.jpg)
Input Scan
Join Scan
Linear Datalog± Scan
Output Scan
Cache*
In-Memory Stream Processing
*an extension point: caching can bein-memory, distributed (e.g., Ehcache), …
![Page 49: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/49.jpg)
Input Scan
Join Scan
Linear Datalog± Scan
Output Scan
Cache Index
In-Memory Stream Processing
Cache*
*an extension point: caching can be
in-memory, distributed (e.g., Ehcache), …
![Page 50: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/50.jpg)
Input Scan
Join Scan
Linear Datalog± Scan
Output Scan
Cache Index
Termination Strategy
In-Memory Stream Processing
*an extension point: caching can be
in-memory, distributed (e.g., Ehcache), …
Cache*
![Page 51: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/51.jpg)
(Person with significantcontrol over a company)
Performance
PVLDB 11(9) 2018
![Page 52: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/52.jpg)
More benchmarks in Luigi Bellomarini, Emanuel Sallinger, Georg Gottlob: The Vadalog System: Datalog-based Reasoning for Knowledge Graphs. PVLDB 11(9) 2018
![Page 53: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/53.jpg)
PAPER ON THE VADALOG LANGUAGE
• Marcelo Arenas, Georg Gottlob, Andreas Pieris: Expressive languages for querying the semantic web.ACM TODS 13:1-45, 2018.
PAPERS ON THE VADALOG SYSTEM
• Luigi Bellomarini, Georg Gottlob, Andreas Pieris, Emanuel Sallinger: Swift Logic for Big Data and Knowledge Graphs.International Joint Conference on Artificial Intelligence (IJCAI) 2017
• Luigi Bellomarini, Emanuel Sallinger, Georg Gottlob: The Vadalog System: Datalog-based Reasoning for Knowledge Graphs.PVLDB 11(9) 2018.
…
![Page 54: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/54.jpg)
Some Applications
with two special partners/customers
![Page 55: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/55.jpg)
1. Company Controlnew approaches to classical problems – when does a company control another company?
2. Close Linksunderstanding whether companies are “too close” in terms of mutual stock participation for different purposes, e.g., for loan granting
3. Detection of Family Businessidentifying families along with their ownerships, i.e., considering the family as the elementary control unit
4. Anonymization of Confidential Data deciding whether a dataset respects complex confidentiality criteria (e.g., ISTAT) before publication and, if not, make it anonymous
5. Hybrid Data Science Pipelineswith different data sources, machine learning frameworks, programming languages, …
Collaboration
Banca d’Italia
Central Bank of Italy
… more applications that we cannot talk about at this point
![Page 56: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/56.jpg)
1. Company Controlnew approaches to classical problems – when does a company control another company?
2. Close Linksunderstanding whether companies are “too close” in terms of mutual stock participation for different purposes, e.g., for loan granting
3. Detection of Family Businessidentifying families along with their ownerships, i.e., considering the family as the elementary control unit
4. Anonymization of Confidential Data deciding whether a dataset respects complex confidentiality criteria (e.g., ISTAT) before publication and, if not, make it anonymous
5. Data Science: Hybrid pipelineswith different data sources, machine learning frameworks, programming languages, …
… more applications that we cannot talk about at this point
Collaboration
Banca d’Italia
Central Bank of Italy
![Page 57: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/57.jpg)
4. Anonymization of Confidential Data
According to ISTAT guidelines,
the survey features are classified
in Direct, Numerical, Related,
Rare, Visible, Traceable, Sensitive
On the variables:
Fiscal code (=SSN) (f), age (a),
region (r), education(e)
We have a statistical survey
about people that needs to
be anonymized
![Page 58: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/58.jpg)
5. Hybrid Data Science Pipelines
Building hybrid data science pipelines,
including Vadalog, Python, Kafka, Flink,
Dgraph, GraphX, ML, …
One approach for interaction between
Jupyter, ML and Vadalog: Python native
driver for Vadalog
Example (next slide):
Company-type classification problem using “domain knowledge”:
Italian SAE-Code
![Page 59: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/59.jpg)
Calculate $HighInc, $MaxInc, $MinInc parameters in Python
and pass them to the Vadalog KG
Nace KG (part):
![Page 60: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/60.jpg)
5. Hybrid Data Science Pipelines
![Page 61: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/61.jpg)
1. Entity Resolution
2. Similarity in Bipartite Graphs
3. Knowledge Graph Support
4. Computing Higher-Level Events and Signals on KG
5. Fact Enrichment and Verificiation on KGs
Collaboration
“ACME”
… more applications that we cannot talk about at this point
![Page 62: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/62.jpg)
1. Entity Resolution
2. Similarity in Bipartite Graphs
3. Knowledge Graph Support
4. Computing Higher-Level Events and Signals on KG
5. Fact Enrichment and Verificiation on KGs
Collaboration
“ACME”
… more applications that we cannot talk about at this point
![Page 63: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/63.jpg)
• Coverage of Internal KG (IKG)Team
(via machine learning)
and Vadalog DeepReason Solution (DR)
• 7,994 company pairs linked by IKG
•16,379 company pairs linked by VADA
•10,413 company pairs identified by VADA as
strongly linked
• Accuracy (1,200 manually inspected company pairs)
1. Entity Resolution
![Page 64: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/64.jpg)
1. EntityBlocking
• Reduce 𝑂(𝑛2) complexity by comparing only entities with similar attributes• Extensive use of text cleaning functions to normalize input values
2. Entity Comparison
• Compare the similarity of every pair of entities for each of their attributes• Extensive us of text similarity functions• Normalize the similarity into a probability 𝑃𝑖, for each attribute 𝑖
3. Probability Computation
• For each pair of entities, compute the overall probability from the individual attribute probabilities 𝑃𝑖 using the Naive Bayes formula
𝑃 =∏𝑃𝑖
∏𝑃𝑖 +∏ 1 − 𝑃𝑖
Entity resolution in three steps
1. Entity Resolution
![Page 65: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/65.jpg)
1. Entity Blocking
key(Id, Key) :-entityName(Id, Name), Key = sim:trim(
sim:removeNonWord(sim:removeDiacritics(sim:toLowerCase(Name)))).
compute a key for blocking by applying cleaning functions
on attributes (e.g. names, addresses, urls, etc.)
block(DId, CId) :-key(DId, Key),key(CId, Key).
establish the pairs of entities to be compared
![Page 66: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/66.jpg)
2. Entity Comparison
compute similarity per attribute (e.g. "name", "address", etc.)
attributeSimilarity(DId, CId, "name", Sim) :-block(DId, CId), entityName(DId, DName), entityName(CId, CName), Sim = sim:mongeElkan(DName, CName).
+ several special rules expressing specific knowledge, e.g. about URL structure (zurich.ibm.com vs almaden.ibm.com)
![Page 67: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/67.jpg)
3. Probability Computation
overallProbability(DId, CId, OverallProbability) :-attributeSimilarity(DId, CId, Att, Prob),ProbProd = prod(Prob),InvProbProd = prod(1 - Prob),OverallProbability = ProbProd / (ProbProd + InvProbProd).
combine similarities (as probabilities) using Naive Bayes
with simple use of aggregates
…or any other ML-based or statistics-based way to combine similarities
![Page 68: Knowledge Graphs and Enterprise AIconferences.cis.umac.mo/icde2019/wp-content/uploads/2019/...Knowledge Graphs as Large World KBs Cyc [Lenat &Guha 1989]: ^comprehensiveontology and](https://reader034.vdocument.in/reader034/viewer/2022050118/5f4f04d1b4d989614c053e39/html5/thumbnails/68.jpg)
Thank You!