Source: https://www.w3.org/Addressing/URL/uri-spec.htmlInitial paper from Berners-Lee: https://www-sop.inria.fr/acacia/cours/essi2006/Scientific%20American_%20Feature%20Article_%20The%20Semantic%20Web_%20May%202001.pdf
The Semantic web, sometimes known as Web 3.0 (not to be confused with Web3), is an extension of the World Wide Web through standards set by the World Wide Web Consortium (W3C). The goal of the Semantic Web is to make Internet data machine-readable.
To enable the encoding of semantics with the data, technologies such as Resource Description Framework (RDF) and Web Ontology Language (OWL) are used. These technologies are used to formally represent metadata.
Advantages of Property Graphs
- Simplicity: Property graphs are simple and quick to set up and use. Knowledge graphs based on property graphs can be an excellent start for new users.
- Easy Navigation: Property graphs are easier to traverse without limitations or standard querying languages.
- Detailed: Properties associated with relationships in property graphs offer more detail about the data entities and their relationships without having to create extra nodes for every detail. The interpretation of the information is left up to the user.
Disadvantages of Property Graphs
- Lack of global IDs: No URIs like in semantic web. Unique identifiers are local to the property graphs and have no meaning to any other database.
- Lack of Interoperability: The lack of standardization (no shared simple schema language like RDF, RDFS, OWL) makes it difficult to share or exchange data with different data stores and prevent graph merging.
- No self-describing data: The concept of embedded or referenced ontology does not exist in labeled property graphs
- Vendor Lock-in: Organizations using a knowledge graph based on property graphs cannot integrate their information across multiple tools or systems. The possibility of being locked into a single property graph vendor is very high.
Advantages of RDF Graphs
- Standardization:All RDF-based knowledge graphs use the same standard framework and formal semantics for storing and representing data along with a standard querying language. Data sharing between RDF data stores on the web is simplified thanks to RDF’s web-native syntax.
- Interoperability: RDF Triple Stores follow a W3C-supported standard that allows interoperability among knowledge graphs. This interoperability allows RDF-based graphs to integrate and exchange information with each other.
- Extensibility: RDF Graphs allow users to add new nodes and relationships, or even substructures, without requiring rebuilding the database.
Disadvantages of RDF Graphs
- Deep Search Complexity: Performing a deep search in large RDF graphs is a complex task as it requires traversing through every relationship.
- Strict Adherence to Standards: All information stored in RDF should be in the form of triples meaning only two objects can be linked at a time, which can be limiting for many use cases.
A Gartner study stated that graph analytics are among the most underleveraged capabilities because organizations don’t understand the contrast and complementary nature of graph insights relative to relational analysis.
For humans and machines, knowledge graphs describe and contextualise together micro and macro business complexity using domain model, enterprise architecture and digital twin
- Alignment and engagement by aligning understandings across businesses on what each keystone concept means and documenting, you improve communication across teams and increase shared understanding of goals and increase the sense of purpose at work of stakeholders.
- Data Governance and Data integration: A knowledge graph helps you
- contextualise your data asset and link it to data knowledge in order to adapt you governance
- link existing data from different silos (organisational, systems, internal/external,...) to make your existing data more valuable and more useful.
- Knowledge aggregation and artificial intelligence: Knowledge graphs accumulate institutional knowledge and expert experience as they grow. This opens the door to more reliable, ethical machine-learning for search and recommendations. KG can serve as curated and governed input to feed generative AI.
- Data science and productisation Knowledge graphs represents avatars of data assets and describe them within context including provenance from system-of-record, lineage through transformation and required product metadata hence ensuring transparency and trust.
Use an ontology to consistently and semantically document and manage
- list of term or values
- taxonomies of terms or values
Use an ontology to consistently and semantically document and manage identifiers for shared entities in order to reuse them seamlessly across humans and machines
Use an ontology (e.g. DCAT) to consistently and semantically make lists of assets (including data sets)
- Inventories: for internal use (not for customers) describing what we have in stock
- Catalogs: for external publication (for customers) to understand all products and/or services available for their consumption
Use an ontology to consistently and semantically document data architecture at conceptual, logical and physical levels.
Use an ontology to consistently and semantically document
- where data comes from (system-of-record or source-of-truth)
- where it goes (target migration system, data product, platform)
- how it is processed or transformed along its journey
Identification, extraction, mapping and disambiguation of PII.
As per GDPR, any person (applicant, employee, supplier...) can ask a company to identify her/his personal information in all systems and request it to be deleted.
After data discovery and extraction, storing and mapping the data in a KG supports the disambiguation of retrieved entities.
Many traditional technologies aren’t designed to connect the dots, so detecting money laundering schemes requires a tremendous amount of laborious effort. Teams of inspectors are burdened with manually going through gobs of data.
Using a graph to depict web pages and their keywords using web scrapping and NLP
- Automatically extract web page structure and content
- Store it in a KG
- Mapping similar entities and create information clusters
- Oppose different content using specific relationships
KPI differ in the contexts of B2B and B2C customers even though the meaning of CLV is identical.
CLV for B2B customers is frequently related to leasing an entire fleet and depends on macro-economic drivers and corporate finances, while B2C customers are privately leasing individual cars that fit to their personal stage of their life. Specific customers are established as instances of the B2B or B2C class. Consequently, such context is highly relevant to derive appropriate business rules to maximize the CLV of the customer base.
“Pasted image 20240125094838.png” could not be found.
Making effective real-time recommendations depends on a data platform that understands the relationships between entities. Only graph technology efficiently tracks these relationships according to user purchase, interactions and reviews.
Use an ontology to consistently and semantically document business building blocks and how they interact together.
Definition: Collection of terms used for a specific purpose or belonging to a specific business domain. Terms can be organised as lists or taxonomies
List
Flat list of terms (graph nodes) with no specific structure.
Taxonomy
A hierarchical of terms (graph nodes) with parent-child relationships (graph edges) between upper-indented (lower) terms.
Many companies currently face challenges caused by demographic changes: In the upcoming years employees will retire more frequently than graduates will join, an effect frequently described as ‘war for talent’. As a qualitative consequence, experience and knowledge will be lost. In such a situation, capturing and formalizing expert knowledge about a certain domain is highly beneficial to retain knowledge in a company while replacing retired employees. The use of ontologies and knowledge graphs therefore has significant advantages in their capability to store and retrieve domain knowledge in an understandable format.
Graphs are designed to express relatedness. Graph technology uncovers patterns that are difficult to detect using traditional representations such as tables.
A graph database can store complex, densely connected access control structures spanning billions of parties and resources. Its richly and variably structured data model supports both hierarchical and non-hierarchical structures, while its extensible property model allows for capturing rich metadata regarding every element in the system.
Most enterprise manufacturers use vendor applications: CRM systems, work management systems, accounts payable, accounts receivable, point of sales systems, and so on. Due to this approach, you need to store and model data as a graph where a native graph stores interconnected master data that’s neither purely linear or hierarchical.
By nature, supply chain management is dynamic, with many moving parts, and where bottlenecks may occur at any given point. The challenge is that the volume and detail of data generated by traditional databases lacks real-time, accurate information processing capabilities.
Definition: A symbolic representation of information using visualization techniques.
A diagram from knowledge graph is a subset of nodes and edges organised topologically in a predefined manner.
A complete URI consists of
- a naming scheme specifier
- followed by a string whose format is a function of the naming scheme.
Scheme
The first element is the name of the scheme separated from the rest of the object by a colon
e.g. "http" in https://www.w3.org/Addressing/URL/uri-spec.html
Path
What follows the colon in a format depending on the scheme. The path is interpreted in a manner dependent on the protocol being used. However, when it contains slashes, these must imply a hierarchical structure.
More specific than URIs and GUPRIs
Global Unique Persistent Resolvable Identifier
- Global: globally global or global for your business?
- Unique: who and how to ensure uniqueness?
- Persistent: even true disasters?
- Resolvable: As per FAIR principle A1?
- Identifier: for machines and for humans?
Need syntactic and semantic governance of GUPRIs
Human understandable or not?
Contains # or /? https://www.w3.org/wiki/HashVsSlash
Examples:
Non-human understandable URI: https://www.rcsb.org/structure/2C0K
Human understandable URI: https://dbpedia.org/page/Basel
If you go for human understandable GUPRI
Questions you need to ask
- Define your domain
- Define prefix and suffix
- Do you need to add a timestamp?
- Do you need versioning?
- Persistent over time: based on a governance for syntax and syntactics
- Unique: once assigned, a URI shall not be reused or deleted
- Unambiguous: not open to more than one interpretation
- Human and machine readable: human can read and understand them and web browsers or other IT application interpret them
- Vendor and system agnostic: defined, assigned and governed by an internal team. We keep the full freedom to operate.
- Indicative about context: rich syntax referring to business context, system-of-record (SoR) and standards. They indicate what it is and what it contains
- Semantic enablers: URI can auto-describe themselves when becoming URL and resolving in a web page describing the thing they uniquely represent
- Link and buffer to external world: the pace data/information/knowledge evolves internally differs from external world. When mapped to external entities, enterprise-specific URIs enable to decouple enterprise DIK ecosystem from real-world. We can decide when to sync.
While data is stored in database, where is the domain knowledge?
The domain knowledge connected to the data (many-to-many) must be stored in an abstract, machine readable, but humanly usable representation -> KG/ontologies
Federated search across different (relevant?) data sources
Data integration
- Model-driven: no hard-coded rules
- Quick and costless to modify
360 views
- Customer 360
- Product 360
- ...
Data-driven business
- To improve the entire value chain using data and insights generated from data analyses
- Need for a high degree of data quality and a well-founded understanding of the context of data and their business semantics are relevant drivers
Usages of KGs
Good use cases for relational DBs
Definitions:
- A formal, explicit specification of a shared conceptualisation.
- A systematic representation of knowledge, usually applied to one specific domain of knowledge/business, that defines concepts (graph nodes) and relationships (graph edges) between them.
- A map displaying concepts (data, information, knowledge , people, facilities, constraints...everything that is important per your concerns) based on conventions (legend for road, rivers, village, mountains... representation).
- It is a digital map that machines can understand so that human can query it.
Data governance cases
Miscellaneous cases
- Storing, filtering and sorting key-value pairs
- Storing, filtering and sorting time series
Knowledge graph technology stack
Fast processing of huge numbers of records when the structure/schema of the data is known ahead of time
Represent complex dependencies
Ask questions, get facts from the graph
Navigate/explore concepts and their relationships
Find and reduce blind spots
Use an ontology to consistently and semantically document and manage
- Data domains
- Data products
- Responsibilities (ownership)
Use an ontology for any "meta" information you need to report on
- FAIR
- Quality
- GDPR
- ...
The Semantic Web Stack illustrates the architecture of the Semantic Web.
The Semantic Web is a collaborative movement led by international standards body the World Wide Web Consortium (W3C). The standard promotes common data formats on the World Wide Web. By encouraging the inclusion of semantic content in web pages, the Semantic Web aims at converting the current web, dominated by unstructured and semi-structured documents into a "web of data". The Semantic Web stack builds on the W3C's Resource Description Framework (RDF).
1730
Switzerland and Sweden
1870
Library classification Dewey
USA
1900
Semantics, ontology and logic Husserl
Austria
1960-1970
Network databases (CODASYL) and semantic networks Quilian et al.
1970-1990
Predicate logic as the foundation of knowledge representation Hayes et al.
1972
Term coined by a linguist Edgar W. Schneider
Austria
Late 1980s
Project called Knowledge Graphs, focusing on the design of semantic networks
1985
1997 onward
The semantic web, RDF, OWL, etc Lassila et al.
Finland
2005
2007
2012
Google introduced their Knowledge Graph incorporating content extracted from indexed web pages. Entity and relationship types associated with this knowledge graph have been further organized using terms from the schema.org vocabulary.
2019
IEEE combined its annual international conferences on "Big Knowledge" and "Data Mining and Intelligent Computing" into the International Conference on Knowledge Graph.
Nowadays
Modern knowledge graphs and graph databases
Second-hand knowledge for humans and machines
- enables to capture knowledge in a multi-dimensional and infinitely-extendable manner
- open-standard; never out dated; always re-usable
- that can be retrieved when needed
A simplified visual/graphical representation of a complex reality
- that satisfied different levels of complexity to be disclosed on-demand
Standards for integration of various proprietary data formats
- quick and flexible: schema-free
- based on a description (model) and not hard-coded rules
Systems thinking is a way of making sense of the complexity of the world by looking at it in terms of wholes and relationships rather than by splitting it down into its parts.
It has been used as a way of exploring and developing effective action in complex contexts, enabling systems change. Systems thinking draws on and contributes to systems theory and the system sciences.
What knowledge graphs are good (and not good) at
History and usage of knowledge graphs
Key concepts about knowledge graphs
Who benefits from KGs?
Benefits of using knowledge graphs
Populating the KG with discrete facts in the format of simple sentences:
Example of text
The forest is composed by high trees. Green leafs growing on tree branches form a thick foliage that spans from the village to the hill like a impenetrable green wall
Equivalent graph
For humans
- by directly experiencing reality via sensors
- by indirectly understanding descriptions of reality via sensors (mostly visual in our society)
For machines
Human programming using mathematical functions and variables
Machine learning (still based on algorithms created by humans)
- supervised:
- using a human-labelled training set of data and a feedback loop for correction
- goal is to predict an (binary) outcome from input data
- e.g. cat or no cat on image
- unsupervised:
- using unlabelled data; no feedback mechanism
- goal is to identify unknown patterns in data based on similarities or relationships human don't see
- e.g. clustering
is hardly
- understandable
- trustable
- FAIR
- findable
- accessible
- interoperable
- reusable
requires extra effort to generate knowledge
- how to know if the data is useful to solve/answer a problem/question
- data scientist spend 80% of their time understanding/preparing the data
Knowledge: context about what questions or problems the information is contributing to answer or solve.
Information: context around data usually answering the 5 Ws: who, what, when, where, why , how. Additional information about FAIR, data productisation, governance... can be added based on the needs.
Data: only common data to be shared between different stakeholders/systems such as master or reference data. Usually no transactional data.
Several definitions. Focus on discrete mathematics.
A graph is a structure amounting to a set of objects in which some pairs of the objects are in some sense "related".
Knowledge is a form of awareness and ability to adapt (to environment).
Can be segmented in
- Fact-based knowledge (or true knowledge) e.g. humans have two eyes
- Opinion-based (or guesswork knowledge) by virtue of justification e.g. I have two eyes, therefore I'm human
- Belief-based knowledge (propositional knowledge) e.g. I'm a good person
A knowledge graph is a knowledge base that uses a graph-structured model to identify object (or actions) and link them together.
A data base is an organised collection of data in a computer memory managed by a database management system (DBMS), the software enabling the management of the database content.
https://en.wikipedia.org/wiki/Database
A knowledge base is an organised collection of knowledge in a computer memory. A knowledge base is reacher than a database. It contains more information.
https://en.wikipedia.org/wiki/Knowledge_base
What is a knowledge graph?
Content
- What is a knowledge graph
- Benefits of using knowledge graphs
- History and usage of knowledge graphs
- Key concepts about knowledge graphs
- What knowledge graphs are good (and not good) at
- Knowledge graph technology stack
- Why adopting knowledge graphs now
Introduction to Knowledge Graphs
Cedric Berger
June 8th, 2024
contact@biomedima.org
Knowledge graphs are universal descriptors of systems of interacting elements.
Knowledge graphs are often used to store interlinked descriptions of entities – objects, events, situations or abstract concepts – while also encoding the semantics underlying the used terminology.
Universal support for description
Example
Populating a Knowledge Graph
Context
Data, Information, Knowledge
Knowledge and graph definitions
Rich context, rich knowledge
A natural pattern
Example of KG usefulness
History of KGs
Triple store (semantic) specificities
Nodes and edges
GUPRIs
Data vs. model
Model components
Good use cases for KGs
System architecture
Knowledge graph solutions
Best for data integration
RDF stands for Resource Description Framework, a World Wide Web Consortium (W3C) standard created originally to model metadata.
Triple Stores store and express information in a sentence-like structure of three; subject-predicate-object, denoted by two nodes connected by a single edge. For e.g. Josh likes bread.
Semantic vs. property graphs
How to feed a knowledge graph
Best for data analytics
Property Graphs, also called Labeled Property Graphs (LPG), are a variant of graph databases where entities and their relationships have associated attributes.
Attributes can be any property that gives details of a data entity or a relationship. Property graphs get their name from their capability to include properties associated with nodes and edges denoted as key-value pairs. For e.g. Mark wrote a book. In this case Mark is a data entity represented as a node in the graph. Its associated key-value pair could be Person: Author.
Property graphs are focused on offering faster querying and extensive storage.
Why adopting knowledge graphs now?
RDF standard
1. Don't miss it
2. It makes your business smarter
3. It boosts your freedom to operate
4. You get a competitive advantage
A KG enables you to do things that you currently don't or poorly do. In particular, KG support data-centric data governance, data science (including AI) and monitoring business value delivery.
A KG can be big or small as you want it to be to add value. Using open standards, it maps and represent inter-dependencies from vision, mission to keystone concepts and expand it into an encyclopedia if you need to.
Our digital world is becoming more complex everyday and we understand less and less of it. Without a proper digital transformation, the daily-growing digital complexity of modern businessess overwhelms human brain's cognition.
Most of your competitors are already using a knowledge graphs at the heart of their businesses. Their graphs can be good, bad, flimsy, or robust, but they have them and use them.
| Differences | RDF Triple Stores | Labeled Property Graphs |
|---|---|---|
| Representation | Entities and relationships are represented in a subject-predicate-object structure. | Entities and relationships have associated attributes represented as key-value pairs. |
| Querying Languages | Standard querying language SPARQL. | Every property graph implementation generally has its own language. |
| Internal Structure | Entities and relationships in RDF do not have any internal structures and are only recognized by their URIs. | Entities and relationships in property graphs have an internal structure that goes beyond labels and includes properties as a part of their identity. |
| Focus | RDFs are focused on offering standardization and interoperability. | Property graphs are focused on data entities to enhance storage and speed up querying. |
| Use Cases | Useful for any use case requiring knowledge graphs with slow-changing datasets. Perfect for scenarios requiring reasoning or inference or where information from other data stores is required. E.g., testing or evaluation. | Useful for any use case requiring large knowledge graphs with dynamic datasets that need deep traversal from time to time, e.g., social graphs. |
No entity exists in complete isolation.
Everything is a part of a system
The purpose of life is to connect, create a network, a system of communicating entities exchanging information”
Wikidata
DBpedia
Public/open
Linked Open Data
CELLAR
Open targets
Pharos
Hetionet
other life science knowledge graphs
Individuals
Organizations
Private
AirBnB
Microsoft
Walmart
Uber
eBay
Generation of new insights
- unforeseen relationships between entities
- inference
More relevant search by association
Humans think associatively but normal search applications index information hierarchically or rank results based on term frequency.