In order to let computers to work for us, they must understand data: not just the grammar and the syntax, but the real meaning of things.
KEES is an architectural design pattern that establishes specific requirements for Semantic Web Applications. Its purpose is to formally describe domain knowledge with the goal of making it tradeable and shareable.
Domain knowledge refers to information about a specific subject (e.g., a range of products, commercial offerings, a social network, etc.). Knowledge domains are cumulative, with no defined limits on their scope or size.
With KEES, you can formalize and license all the components necessary to construct a knowledge domain. This includes:
Both machines and humans can leverage this know-how to enhance their knowledge.
For more details, refer to the KEES presentation slides.”
Data, as per KEES, encompasses any literal like words, numbers, or generally, any sequence of symbols. For instance, strings such as
http://LinkedData.Center exemplify data. Typically, data is linked with a data type that defines constraints on the sequence of symbols composing the data. For instance, the number
123, the float
33.22, or the URI
http://LinkedData.Center each possess their respective data types, which specify rules regarding the permissible symbols sequence. Importantly, the data type does not inherently convey the meaning of the data itself.
Information, as defined in KEES (also known as facts), is characterized as data with meaning. This meaning can either be inferred from the context where the data is situated or explicitly defined. KEES adopts the RDF standard to delineate information through a triplet of three elements, forming an RDF statement, commonly referred to as a triple. This triple comprises a subject, a predicate, and an object. Indeed, a triple can be visualized and represented as an undirected labeled graph, and it can be stored within a Graph Store. This store serves as a repository for organizing and maintaining collections of these graph-based representations, allowing efficient storage and retrieval of triples.
KEES defines knowledge as a graph of linked information (i.e. linked data). This graph is possible because, in RDF, any URI can be both the object of a triple and the subject of another one or even the predicate for another.
In KEES, a Knowledge Graph is characterized as a Graph Store with a purpose. It represents a collection of interconnected information (i.e. Linked Data) organized in a graph-like structure that’s designed specifically to be queried, providing answers to specific questions within its domain of related information.
From a theoretical perspective, a knowledge graph comprises information (facts), a formal logic system for knowledge representation, incorporates the Open-world assumption, and employs an inference engine to demonstrate theorems.
In a knowledge graph, information is partitioned into two datasets: TBOX and ABOX. ABOX statements delineate facts, while TBOX statements define the terms used to qualify the meaning of those facts. If you’re familiar with the object-oriented paradigm, TBOX statements can be likened to associations with classes, whereas ABOX statements are linked to individual class instances.
TBOX statements are typically more enduring within a knowledge graph and are often organized into ontologies, describing specific knowledge domains such as business entities, people, goods, friendships, offerings, geocoding, and more. These statements are expressed using an ontology language that offers a formal semantic, like the W3C OWL.
ABOX statements, on the other hand, are related to instances of classes defined by TBOX statements. They possess a much more dynamic nature and are populated from datasets available on the web or through reasoning processes.
Theorems within a knowledge graph can manifest as rules or axioms. A rule signifies a generalized inference that establishes a logical correlation between propositions. Conversely, an axiom embodies a rule accomplished through entailment that is inferred by the semantics of existing factual information. Axioms and rules can implement deductive and abductive reasoning.
For instance, an axiom could be articulated using OWL (Web Ontology Language) and demonstrated through an OWL reasoner. Alternatively, a rule could be formulated using SPARQL QUERY CONSTRUCT and materialized with SPARQL UPDATE operations.
The Language Profile (also referred to as semantic Application Profile) forms the section of the TBOX comprising all terms recognized by rules and employed to comprehend and respond to inquiries. This profile defines the scope and vocabulary utilized within a Semantic Web Application, outlining the terms essential for understanding and generating responses within its semantic framework.
The term KEES Agent refers to a processor capable of understanding the KEES language profile and adhering to the KEES protocols. There are seven types of KEES agents:
Trust holds significant importance in KEES due to the Open-world assumption and the inherent nature of RDF, which permits the amalgamation of diverse information, even when it might be incoherent or falsified.
In the context of KEES, trust serves as metadata linked to each RDF statement, ranging in value from 0 (indicating no trust) to 1 (indicating complete trust). Notably, this concept of trust extends to both ABOX and TBOX statements within the knowledge graph.
Moreover, the trust assigned to an inferred fact is intricately tied to the trust levels associated with all the involved facts and rules contributing to its derivation. Therefore, the trustworthiness of an inferred fact is contingent upon the collective trust levels of the underlying information and the processes involved in its generation within the knowledge graph.
The trust value of 0 or 1 signifies certainty in a fact. This principle extends to TBOX statements as well.
Note that the trust of an inferred fact MUST be less or equal to the minimum trust of the involved statements.
Inferred facts can be derived by the following reasoning type:
For instance, suppose that an axiom in your knowledge graph TBOX states that a property “:hasMom” has a cardinality of 1 (i.e. every person has just one “mom”), your knowledge graph could also contain two different facts (:jack :hasMom :Mary) and (:jack :hasMom :Giulia), perhaps extracted from different data sources. In order to decide who is Jack’s mom, you need trust in your information.
If you are sure about the veracity of all data in the knowledge graph, you can deduct that: Mary and Giulia are two names for the same person. If you are not so sure, you have two possibilities: deduct that the one data source is wrong, so you have to choose the most trusted statement concerning some criteria (even casually if both statements have the same trust rank) or change the axiom in TBOX, allowing a person to have more than one mom. In any case, you need to get an idea about your trust in each statement, both in ABOX and in TBOX, in the knowledge graph. At least you want to know the provenance and all metadata of all information in your knowledge graph because the trust in a single data often derives from the trust of its source or in the creator of the data source.
- fact 1:
:jack :hasMom :Mary
- fact 2:
:jack :hasMom :Giulia
- axiom :
?restriction owl:qualifiedCardinality=1; owl:onProperty :hasMom
The trust helps to solve potential conflict according to this table:
trust in fact 1 trust in fact 2 trust in axiom derived facts derived fact trust reasoning type k <= 1 k <= 1 k <= 1 new fact (:Mary owl:sameAs :Giulia) k deduction x < 1 y > x y<=k<=1 fact1 wasInvalidatedBy (axiom + fact 2) <=y abduction x < 1 y < x x<=k<=1 fact2 wasInvalidatedBy (axiom + fact 1) <=x abduction k k z<=k<=1 axiom wasInvalidatedBy ( fact 1 + fact 2) <=z induction j < z j < z z<=1 ( fact 1 + fact 2) wasInvalidatedBy axiom z free will
Note that the last row in the table is a just subjective illogic paranoic response, you could also decide to:
- invalidate the axiom (i.e. take the risk of fake data)
- randomly invalidate one of the conflicting facts (i.e, take the risk of wrong rule )
- derive that :jack is not an human and add a restriction on :hasMom property domain ( creative thinking )
- … Resolving the trust conflict it depends from your risk appetite and from your free will.
The processing of KEES data involves a sequence of phases, referred to as windows, executed by KEES agents within a knowledge graph.ù
The sequence of windows, known as the KEES cycle, constitutes a continuous integration workflow. It starts upon a triggered change in learned facts and ends in a fully populated knowledge base ready for application use. The KESS cycle windows are:
Steps 2, 3, and 4 can be iterated as needed for the processing of KEES data within the knowledge graph.
The duration between the detection of new data and the creation of a new stable version of the knowledge graph is termed Knowledge Graph Convergence time or just convergence. It’s important to note that the convergence is always > 0, defining the duration for the entire KEES data processing, which can be conceptualized as a big ETL (Extract, Transform, Load) process.
LinkedData.Center SDaaS product is a commercial early and partial implementation of KEES specifications (also available the Open Source community version)
A great way to contribute to the site is to create an issue on GitHub when you encounter a problem or something. We always appreciate it. You can also edit the code by yourself and create a pull request.
All stuff here in the Creative Common (unless otherwise noted).