Slides: Knowledge Graphs vs. Property Graphs

Knowledge Graphs vs. Property
Graphs — Similarities, Differences
and Some Guidance on Capabilities

TOPQUADRANT COMPANY
FOUNDATION
▪ TopQuadrant was founded in 2001
▪ Strong commitment to standards-based approaches to data semantics
MISSION
▪ Empower people and drive results — by making enterprise information meaningful
FOCUS
▪ Provide comprehensive data governance solutions using knowledge graph technologies
TOPQUADRANT COMPANY

© Copyright 2020 TopQuadrant Inc. Slide 3
I. A brief overview of:
– Property Graphs
– Knowledge Graphs
II. Differences in Terminology and Capabilities of these models
III. Strengths and Limitations of each approach
IV. Why Knowledge Graphs provide a strong foundation for data
centric enterprise applications
– E.g. enterprise data governance and metadata management
▪ Q & A
Today’s Agenda
Irene Polikoff

Some Definitions of a Knowledge Graph
▪ A knowledge graph (i) mainly describes real world entities and their
interrelations, organized in a graph, (ii) defines possible classes and relations of
entities in a schema, (iii) allows for potentially interrelating arbitrary entities
with each other and (iv) covers various topical domains.
H. Paulheim. Knowledge Graph Refinement: A Survey of Approaches and Evaluation Methods.
Semantic Web Journal, 1–20, 2016
▪ A knowledge graph acquires and integrates information into an ontology
and applies a reasoner to derive new knowledge.
Lisa Ehrlinger and Wolfram Wöß, Towards a Definition of Knowledge Graphs,
Institute for Application Oriented Knowledge Processing at SEMANTiCS 2016, http://ceur-ws.org/Vol-1695/paper4.pdf
▪ A knowledge graph is a knowledge base that uses a graph-structured data
model or topology to integrate knowledge and data. Knowledge graphs are often
used to store interlinked descriptions of entities — real-world objects, events,
situations or abstract concepts, including their semantics.
Wikipedia

TopQuadrant: What Are Knowledge Graphs?
▪ A Knowledge Graph represents a knowledge domain
– With “Active Models” that can be consulted at run-time
▪ It represents all information as a graph
– A network of nodes and links
– Not tables of rows and columns
▪ It represents facts (data) and models (metadata) in the
same way – as part of a graph
– Rich rules and inferencing
▪ It is based on open standards, from top to bottom
– Readily connects to knowledge in private and public clouds
A knowledge graph, with its Active Models, grows and evolves over time.

Knowledge Graph Recipe
Vehicle
Car
Organization
Year
Design/Model
brand
type
type type
type
BMW X2
BMW
2019
BMW
228
Ontology
+ Instance Data
----------------------------
Knowledge Graph
*Blueprint *Actuals
An Ontology is a data model (schema) with classes, properties, and rules - a precise blueprint for data

Property Graphs
aka Labeled Property Graphs
▪ Caveat: There is no standard property graph data model
– Each implementation of a Property Graph is, thus, somewhat different
– We will focus on the common aspects
▪ In general, the Property Graphs data model consists of three elements:
– Nodes – entities in a graph
– Edges – direct links aka relationships between nodes
– Properties – key value pairs that can be associated with either nodes or edges

Property Graph by Example

RDF Knowledge Graph by Example
56853
http://www.wikidata.org/entity/Q2263
wikidata:

56853

Property Graph vs RDF Knowledge Graph
Property Graph Knowledge Graph
IDs are internal to a graph database, user has no control
over them.
IDs are global – URIs, meant to be under users control to
enable combining different graphs
Properties are literal values. They are fundamentally
different from nodes and relationships.
Canonical structure. Everything is stored as nodes and
links connecting them. A literal value is a node like any
other. Property is any link – to a resource or a literal.

“Schema” as part of a Knowledge Graph

More “Schema”
sh:maxCount
1

Standards-based Ontologies and Vocabularies are
Knowledge Graphs Increasingly Used across Industries
Key enablers of transparency, interoperability
and the emerging “cognitive enterprise”
These ontologies uniquely identify and describe things and
relationships in healthcare, web search services,
manufacturing, financial services, …
*

IDs are internal to a graph database, user has no
control over them.
IDs are global – URIs, meant to be under user’s control to
enable combining different graphs.
Canonical structure. Everything is stored as nodes and
links connecting them. A literal value is a node like any
other. Property is any link - to a resource or a literal.
Schema (semantics of data) is not a part of the graph. Rich schemas, including rules are a part of the graph.

RDF Knowledge Graphs:
How to add information to triples

over them.
Canonical structure. Everything is stored as nodes and links
connecting them. A literal value is a node like any other.
Property is any link - to a resource or a literal.
Each relationship (link or arch) uniquely identifies “node
– link – node” combination. Relationships can be
annotated with additional facts, but properties can’t be.
IDs of properties (links) are re-used. Thus, they do not
uniquely identify a “node – link - node” combination that
uses it. There is a way to give these triples identity. Any
triple can be annotated with additional facts.

Property Graphs:
How to add information to a “property” value
Can we say what year is the population for?

type: POPULATION_SIZE type: POPULATION_SIZE
label: Population
label: Population
131 132“Size” 58811
“Year” 2018
“Size” 56853
“Year” 2010
Property Graphs:
How to add information to a “property” value
Can’t be done. Must turn a “property” into a relationship

RDF Knowledge Graph: How to add
information to a triple with a literal object
Or, as it more commonly said, “to a property with a literal value”
ex:year
2010
ex:100058811 56853

Property Graph: How to use a node as a
“property” value
Let’s say we want to capture information
about Ben Bradlee so we want to turn
the value of “role” into a node.

Property Graph: How to use a node as a
“property” value
Must remove the edge
annotation and create a
relationship to a node
labeled Person.

RDF Knowledge Graph: How to turn a
literal value into a resource reference

over them.
Each relationship (link or arch) uniquely identifies “node –
link – node” combination. Relationships can be annotated
with additional facts, but properties can’t be.
IDs of properties (links) are re-used. Thus, they do not uniquely
identify a “node – link - node” combination that uses it. There is
a way to give these triples identity. Any triple can be annotated
with additional facts.
Changes in the graph design require restructure/re-load of
the data and changes to all impacted queries.
Graphs can evolve and changes in the design can often be
done with minimal impact on existing data and queries.
Product-specific query languages, variants of Cypher,
increasingly, GraphQL support, introspection not integrated
Query standard – SPARQL. Increasingly, GraphQL support. In
EDG: introspection and auto-generation of GraphQL Schemas.

wikidata:Q2263
rdf:type entont:Actor ;
rdf:type entont:Director ;
entont:actedIn entertainment_data:ThePost
[[ entont:role “Ben Bradlee” ]] ;
schema:birthDate "1956-07-09"^^xsd:date ;
entont:directed entertainment_data:ALeagueOfTheirOwn ;
schema:familyName "Tom" @en;
schema:givenName "Hanks" @en;
rdfs:label "Tom Hanks" @en; .
Example of RDF Serialization
@prefix entertainment_data: <http://example.org/data-graphs/Entertainment_Data#> .
@prefix entont: <http://example.org/ontologies/Entertainment#> .
@prefix wikidata: <http://www.wikidata.org/entity/> .
@prefix schema: <http://schema.org/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
Subject Node
Object NodeEdge

over them.
Each relationship (link or arch) uniquely identifies “node –
link – node” combination. Relationships can be annotated
with additional facts, but properties can’t be.
IDs of properties (links) are re-used. Thus, they do not uniquely
identify a “node – link - node” combination that uses it. There is
a way to give these triples identity. Any triple can be annotated
with additional facts.
Changes in the graph design require restructure/re-load of
the data and changes to all impacted queries.
Graphs can evolve organically. Changes in the design can often
be accomplished with minimal impact on existing data and
queries.
Product-specific query languages, variants of Cypher,
increasingly, GraphQL support, introspection not integrated
Query standard – SPARQL. Increasingly, GraphQL support. In
EDG: introspection and auto-generation of GraphQL Schemas.
No standard serialization for export.
Standard serializations supported by all products –
RDF/XML, Turtle, N3 and JSON-LD formats.

Knowledge Graphs are Self-Organizing –
they can be composed
Data
Element
Dataset Data
Center
Location
element of located in
KG1 KG3
located in
CMS
CMS
Cloud
Service
hosted by
KG2
App.
Server
VM
resourcedrunning on
stored in

Knowledge Graphs are Self-Organizing –
they can be composed
With consistent Identifiers for both Nodes and Links,
Graph queries and rules can give insights and make new knowledge.
Data
Element
Dataset
Data
Center
Location
element of located in
KG1 KG3
located in
CMSCMS Cloud
Service
hosted by
KG2
App.
Server
VM
resourcedrunning onstored in regioned in

IDs are internal to a graph database, user has no control over them. IDs are global – URIs, meant to be under user’s control to enable
combining different graphs.
Properties are literal values. They are fundamentally different from
nodes and relationships.
Canonical structure. Everything is stored as nodes and links connecting
them. A literal value is a node as any other. Property is any link - to a
resource or a literal.
Each relationship (link or arch) uniquely identifies “node – link –
node” combination. Relationships can be annotated with additional
facts, but properties can’t be.
IDs of properties (links) are re-used. Thus, they do not uniquely identify
a “node – link - node” combination that uses it. There is a way to give
these triples identity. Any triple can be annotated with additional facts.
Changes in the graph design require restructure/re-load of the data
and changes to all impacted queries.
Graphs can evolve organically. Changes in the design can often be
accomplished with minimal impact on existing data and queries.
Product-specific query languages, variants of Cypher, increasingly,
GraphQL support, introspection not integrated.
Query standard – SPARQL. Increasingly, GraphQL support. In EDG:
introspection and auto-generation of GraphQL Schemas.
No standard serialization for export. Standard serializations supported by all products – RDF/XML, Turtle, N3
and JSON-LD formats.
No partitioning of graphs nor merging or connectivity
across graphs.
Named graphs as a partitioning approach. Graphs are “self-
composing”, can be connected and merged.

Knowledge Graphs support the move to
Data Centric
Systems and Architectures

Data Centric vs Application Centric
Application Centric
Data is a second-class citizen, and the
application is the main thing
Data models are not self describing, can’t be
understood without understanding
application’s logic
All business rules and semantics are hard
coded in the applications
Data models are very rigid
Changing application’s data model is
expensive – requires changes to the
application code due to tight coupling
One data model per application
Data sharing is an afterthought
Data Centric
Data is the main asset, applications are
“model-driven”
Data is self-describing and can stand on its
own
Data can be “active” with accessible
semantics, business rules and even
executable code
Data models are flexible and evolvable
Changing data models is inexpensive and may
not require application changes
Multiple applications may share some parts of
a data model, extending it as needed
Data sharing is “built in”

Why Self Describing Data is Important?
Open Data Example: Road Safety - Vehicles by Make and Model
We see the numbers, but what do they mean?
Data is only as good as its metadata

Metadata adds Context & Definition –
Making Data sharable and increasing its Value
39
First Name Last Name Company City
Year
Purchased
Joe Smith Komputers R Us New York 1970
Mary Jones The Lord’s Store London 1999
Proful Bishwal The Lady’s Store Mumbai 1998
Ming Lee My Favorite Store Beijing 2001
Customer Definition
Last Name represents the surname or family name of
an individual.
Business Rules
In the Chinese market, family name is listed first in
salutations.
Format VARCHAR(30)
Abbreviation LNAME
Required YES
More …
Numerous technical & business metadata including
security, privacy, permissible values, etc.
Is this the city where the customer lives
or where the store is located?
• If metadata is not readily available, data can’t be understood.
• If metadata is kept separate from data, applications can’t access it.
• Further, data will change and metadata will get stale.
• If metadata is captured in a proprietary way, it is not open.
Key Facts

TopBraid EDG – Composition of Knowledge Graphs
to Support Adaptive Data Governance
Business Logic
UX/UI
Access Control
Compose UI
Behavior,
Queries,
Rules and
Constraints
Enterprise
Assets
Technical
Assets
Big Data
Assets
Data Assets
Glossaries
Terminologies
Governance
Model
DataSchema
Compose
Data and
Schema
Over 300 of pre-defined Asset Types

How Applications can be Model Driven –
an example from TopBraid EDG
Models drive the User Experience (UX) i.e., what fields will appear on an (edit, search or browse) form,
possible values, order of the fields, their division into sections and other relevant information.
edg:FileAsset
rdf:type edg:AssetClass ;
dash:abstract "false"^^xsd:boolean ;
rdfs:label "File Asset" ;
rdfs:subClassOf edg:AccessControllable ;
rdfs:subClassOf edg:ComplianceAspect ;
rdfs:subClassOf edg:Identifiable ;
⋮ ⋮
sh:property edg:File-formatType ;
sh:property edg:File-name ;
sh:property edg:File-residesIn ;
sh:property edg:File-sourceLink ;
.
edg:File-formatType
rdf:type sh:PropertyShape ;
sh:path edg:format ;
sh: description “Specifies the format of a file, or dataset, using an enumeration of values”
sh:class edg:FormatType ;
.
edg:File-name
sh:path edg:name ;
sh:datatype xsd:string ;
.
edg:File-residesIn
sh:path edg:residesIn ;
sh:class edg:FileSystem ;.
An active model (ontology)
describing a File Asset as it is
captured in a Knowledge Graph –
open to access at run time, sharable,
making data self describing – for
people and machines
A form for capturing data about files – if a
model changes, a form automatically
adjusts.
The same form can display information
about other assets – whatever the model
says, it will display and will let you edit
according to the definitions in the model
We know the meaning of each data value
because it is described in the ontology that
is as accessible as the data

Some Limitations of Property Graphs
You are Likely to Face
▪ No Capture of Schema in a Graph
▪ Minimal Support for Validation and Data Integrity
▪ No Capture of Rich Rules in a Graph
▪ No Support for Inheritance and Inference in a Graph
▪ No Globally Unique Identifiers
▪ No Resolvable Identifiers
▪ No Inherent Connectivity Across Graphs
▪ Difficulties with Graph Evolvability

Steps for Transitioning from Property
Graphs to a Knowledge Graph
1. Get your data out – “as-is”
– It is fairly easy to generate one of the RDF standard serializations from a property
graph. In fact, Neo4J offers a library for doing this
2. Create a model for your data
– You will not be able to get the semantics of the data; this is due to the fact that
the data model only exists in your initial design sketches and, partially, within
Cypher queries and programs.
– TopBraid EDG can introspect the data to help you generate the initial model

Steps for Transitioning from Property
Graphs to a Knowledge Graph
3. Iterate
– The initial model and data is likely to benefit from some adjustment as the
structure of the graph data may be influenced by the specific limitations of the
property graph data model and optimizations that were required due to its
architecture
4. Migrate application code and queries
– Take advantage of the data validation, reasoning and model-based introspection
capabilities of the knowledge graphs – you will end up with smaller and more
maintainable code base
– If you already use GraphQL, this will likely simplify the migration

Summary of Key Take Away Points
▪ Increasingly graphs are seen as an excellent choice for data / metadata representation
and management - for the data centric enterprise
– Flexibility is one strong driver: heterogeneous data, integrating new data sources, analytics, and metadata
management and governance all require flexibility and evolvability
▪ The two main graph data models are: Property Graphs and Knowledge (RDF) Graphs
▪ Property Graphs are offered in mature implementations (e.g. Neo4J), that are easy to get started with
– but there are fundamental limitations that are not addressed in the design of property graphs
– E.g., Property Graphs are not self-describing; the meaning of the data they store is not a part of a graph
▪ Knowledge Graphs support the move to Data Centric systems and architectures
– by offering robust capabilities that support flexibility and evolvability
– E.g., Knowledge Graphs capture not only data, but the meaning or semantics of data,
including rich constraints and highly expressive rules
– In the past, implementations of Knowledge Graphs were considered to be more academic, but this has
changed with products like TopBraid EDG
▪ It is possible to move from a Property Graph to a Knowledge Graph in logical steps
– Get your data out – “as-is”; Create a model for your data; iterate and optimize …

… Questions?
Thank You !

Benefits of a Knowledge Graph based
Platform for Data Governance 2.0
As an enterprise knowledge graph infrastructure, TopBraid EDG supports
Data Governance 2.0 and applications of AI / ML
TopBraid Enterprise Data Governance (EDG):
▪ Is flexible and extensible, based on standards
▪ Integrates reasoning and machine learning
▪ Enables people (UI) and software (APIs/web
services) to view, follow and query
▪ Bridges data and metadata “silos” for a
seamless data governance
▪ Delivers Knowledge-driven data governance

To Learn More about TopBraid EDG and Knowledge Graphs:
EDG Product Info:
▪ TopBraid Enterprise Data Governance (TopBraid EDG)
(https://www.topquadrant.com/products/topbraid-enterprise-data-governance/)
Contact us: at info@topquadrant.com to:
▪ Discuss data governance and knowledge graphs
▪ Request a more targeted demo of TopBraid EDG
▪ Ask for a free EDG evaluation account

To Learn More about TopBraid EDG and Knowledge Graphs:
EDG Product Info:
▪ TopBraid Enterprise Data Governance (TopBraid EDG)
(https://www.topquadrant.com/products/topbraid-enterprise-data-governance/)
What Are Knowledge Graphs?
A Knowledge Graph turns Data into Knowledge
(https://www.topquadrant.com/technology/knowledge-graphs/)
Contact us: at info@topquadrant.com to:
▪ Discuss data governance and knowledge graphs
▪ Request a more targeted demo of TopBraid EDG
▪ Ask for a free EDG evaluation account

More Resources ...
More Webinar Recordings, Slides, Q&A:
▪ https://www.topquadrant.com/knowledge-assets/topquadrant-
webinars/
Short Videos:
▪ TopBraid EDG “Quick Grok” Videos
https://www.topquadrant.com/knowledge-assets/videos/
▪ TopBraid EDG Animated Video
https://www.topquadrant.com/project/edg_agile_modular/
Blog:
▪ https://www.topquadrant.com/the-semantic-ecosystems-journal/
Data Governance White Papers
▪ https://www.topquadrant.com/knowledge-assets/whitepapers/

Slides: Knowledge Graphs vs. Property Graphs

More Related Content

What's hot (20)

Similar to Slides: Knowledge Graphs vs. Property Graphs (20)

More from DATAVERSITY (20)

Recently uploaded (20)

Slides: Knowledge Graphs vs. Property Graphs