Content uploaded by Lyubomir Blagoev
Author content
All content in this area was uploaded by Lyubomir Blagoev on Jun 25, 2022
Content may be subject to copyright.
Unified Knowledge Graph Page 1
Inato Ltd, Sofia, 2022
Unified Knowledge Graph
Lyubomir Blagoev, INATO Ltd, lyuboblagoev@gmail.com
Tihomir Blagoev, INATO Ltd, tihomir.blagoev87@gmail.com
Abstract
The creation of software systems with an autonomous behavior needs
powerful tools for semantics presentation in the Knowledge graph (KG)
environment. At the same time, it is necessary to create preconditions for
effective processing of semantics through navigation in KG. This can be
achieved with unification of the KG-construction. The article describes an
approach for such unification as a part of an overall vision for the presentation
of the semantics. The unification solves the targeted problems, but new
questions arise and goals concerning further development, which are
presented.
Key words: Knowledge graph, Semantics presentation, AI
Introduction
Our team in Inato Ltd has extensive experience in software systems building both in the field of
administrative activity and in production. Developing further our systems, we directed our efforts
towards creation of capabilities for our system concerning learning, collaboration between
systems and with humans and decision making.
We understood that this needs powerful tools for semantics presentation. At that time, we had
created an appropriate architecture for such presentation- the Semantic Network Based
Architecture (SNBA) [1] and a vision for that- Semantics representation regarding establishing and
maintaining a semantic Interoperability in the e-Governance's environment [2]. Those theoretical
developments had allowed us to create a Unified Platform for Innovations (UPI).
So we had at our disposal a vision and the tools to create every kind of semantic constructions
whit the tools of the UPI-environment and in the condition of no limitations of the Semantic network.
Very soon we understood that the lack of limitations actually did not help us. Then it turned out
that we should have had some kind of semantic framework, which many authors are calling
Knowledge graph (KG).
Most of our ideas for creating the algorithms concerning autonomous behavior need navigation in
a semantic construction in the KG-environment. This in itself makes us clean our vision of the KG
itself as a semantic construction in the environment of the Semantic Network.
Knowledge graph as a semantic construction
We have reviewed many (the list is too long to fit in this material) constructions of Knowledge
graphs but we have not found KG-construction based on models and their instances existing in
common environment. This "common existence" is the main feature of our UPI-environment
1
where we are creating our present systems and where we will create our future systems with
autonomous behavior.
In the UPI-environment the navigation is independent of the kind of KG-nodes, that is whether
they are models or instances of models. The same independence can be pointed out for the arcs
as well. Analyzing the simplest KG-construction (two nodes connected with an arc), we discovered
important features of semantic interactions in this construction.
For example, at first glance the KG-construction of the semantics in the expressions "The cat is
white" and in "John is a father of Jim" are the same- two nodes connected with arc. But we
1
This feature is in fact a feature of the SNBA [1] architecture of the UPI-environment.
Unified Knowledge Graph Page 2
Inato Ltd, Sofia, 2022
analyzed those constructions from the point of view of semantics transfer and have discovered an
important difference. This made us work towards a new vision of KG-construction as a further
development of our general vision [2] for semantics presentation. So we came up with the idea for
unifying the KG-construction which has led to the definition of a new basic semantic constructions
as follows::
1. Semantics/Meaning
Internal, logical content, comprehended by our mind; the meaning of a word or phrase given
2. Atomic semantics
That’s indivisible semantics, which cannot be decomposed into other components. The atomic
semantics could be reviewed as a basis for creating a semantics which upgrades by itself.
That’s why the atomic semantics can be defined as a basic or main semantics.
3. Single semantics
Single semantics is composed of the main semantics and an additional semantics. What is
specific about the Single semantics is that the context of the main semantics refines its
complementary semantics.
The additional semantics in the single semantics is named its content. The presenting of the
additional semantics can not only be in text form, but also in any other form, which can be
perceived by humans directly or by using suitable mechanical means.
4. Compound semantics
Compound semantics is composed of main semantics and content, which is composed by
other compound semantics, which can be either single or compound. Semantics that is
presented with compositions of other single semantics or compound semantics,
complementing compound semantics, is called content.
5. Data
Data is a common construct to present the single and compound semantics, which consists of
main meaning and meaning presented with its content.
6. (Information) object
Information object is a set of data that can be created, destroyed or identified as a whole. In
many cases instead of “information object” it is often used just as “object”.
7. Model
A set of data describing other data or objects can be defined as a model. The models are
presented in the SNBA environment as objects. All semantic constructions in the SNBA
environment begin from one base model- "model for models", which describes himself.
8. Instance (of the model)
Data or object, created according to a model, is “instance” of that model.
The connection between model and instance can be presented through defining straight and
reversed problem:
A) Straight problem: To create a set of data or object according to a given model
B) Reverse problem: To recognize the model, looking at a certain data or object
For a text presentation of the connection between the model and its instance in SNBA
environment we use the suggested in [1] designation- Instance[Model].
9. Class of objects
A Class of objects consists of objects which have a same compulsory set of data, which can
be a subset of the whole amount of data in the object. The objects in a certain class can be
with different models. But the objects with common model form a corresponding class too.
We accepted the expression <name of class>.co as a way of notation of the class of
objects with a name "name of class".
If the compulsory set of data of the AAA.co consists of the compulsory set of data of the
BBB.co, it means that AAA.co is a subclass of BBB.co.
Unified Knowledge Graph Page 3
Inato Ltd, Sofia, 2022
Every class of objects is represented in SNBA environment with corresponding Object[Class
of objects]. The main information which this object supports is the compulsory set of data
contained in the objects from the represented class.
10. Object[Term]
The purpose of the objects which are instances of Term[Model] is to represent atomic
semantics. Those kinds of objects are part from Term.co.
11. Object[Value]
The purpose of the objects which are instances of Value[Model] is to represent the single
semantics as an xml-construction, which content may be structured as well as unstructured.
Those kinds of objects are part from Value.co.
12. Object]Nomenclature]
The purpose of the objects which are instances of Nomenclature[Model] is to represent a type
of data, analogical by construct to data type “Value”, but with their content pointing out the
main semantics/meaning of other data or atomic semantics. The possible versions of the
content of a concrete Object[Nomenclature] are pointed out in the model.
Those kinds of objects are part from Nomenclature.co.
13. Object[Segment]
The purpose of the objects which are instances of Segment[Model] is to represent as an xml-
construction a compound semantics, which content is constructed by other Compound or
Single data.
Those kinds of objects are part from Segment.co.
14. SNBA nodes.co
All of nodes in SNBA environment are objects belonging to SNBA nodes.co.
15. SNBA arcs.co
All of arcs in SNBA environment are objects belonging to SNBA arcs.co.
The nodes and arcs in SNBA environment are objects, but in practice the word “object” usually
is applicated only for nodes in the graphs. To escape a possible confusion in the SNBA
environment we continue to call graph-nodes just "objects" and when it comes to arcs, we
explicitly remark that it comes to arc-objects.
16. SNBA environment
SNBA environment consists of SNBA nodes.co and SNBA arcs.co. This description can be
accepted as a more precise definition of our Semantic Network Based Architecture [1] - SNBA.
17. SNBA Knowledge graph
In SNBA environment we define Knowledge Graph (KG) as a set of interconnected semantic
constructions.
Here is the place to clarify that, in essence, SNBA-KG is a Property Graph, as is defined in [3],
[4], [5] and others. The difference is in the clarification that the arcs and nodes are defined as
objects, according to the definition of an information object in the SNBA-environment. Other
differences concern the way of arcs identification, which we introduce to achieve a unification of
the SNBA-KG, as is defined further.
18. SNBA Unified Knowledge graph (uKG)
As a main subject of unification, we determine the SNBA arcs, because they play a major role
in semantics processing using navigation in the space of a KG. This role is expressed in the
way of exchange of semantics between two nodes of a KG connected by an arc. We
discovered two way of semantics exchange:
1) Semantics transfer from source node to target node
As a result from this transfer, the target node/object receives a new, unknown kind of
feature, not defined in this model. This means that the accepted semantics modifies
the model of the receiving node/object.
Such kind of connectedness is used for example in graph construction for the
representation of a list of objects. The construction has a source node with arcs
Unified Knowledge Graph Page 4
Inato Ltd, Sofia, 2022
pointing out the objects in the list. Usually, such list is a part of some classification
because of that we call the type of arcs in this construction "classification arcs".
2) Semantics transfer from target node to source node
As a result, from this transfer, the source node/object receives a characteristic that is
included in its model, or is known as a part from the model. This means that the
accepted semantics does not modify the model of the receiving node/object.
The same result can be achieved if the commented transfer of semantics is replaced
by inclusion in the receiving object of a property representing the same characteristic.
This determines the replacement of the property-representation of semantics with
graph-representation as auxiliary, not as basic, and it can be ignored in navigation.
The division of the arcs in the KG into arcs with two main models is a fundamental prerequisite
for processing the semantics in the graph by navigating its topology. This separates the
semantics embedded in the nodes of the graph from the semantics represented by its
topology.
In order to introduce such unification in the SNBA KG we use two models of arcs- Arc-
objects[Classification] and Arc-objects[Feature].
However, reducing the arc models in uKG to two types solves the problem with navigation
when processing the semantics, but raises another problem. When using classification arcs,
this problem boils down to distinguishing arcs belonging to different semantic constructions,
while originating from the same node. And when characteristic arcs are used, there is a need
to indicate with a concrete arc, which characteristic is assigned a meaning.
The both cases can be summarized as processes of specialization of arcs in uKG. Such
specialization of arcs is achieved by adding a Clarifying concept[Value] to the data set in them.
The content of the added data is the identifier of definition of concept or data, which conceptual
meaning is used as a marker for the arcs specialization.
The presence itself of an object in a given ontology sometimes needs additional quantitative
clarification. This is achieved in uKG by adding a Clarifying data[Value] to the data set in the
arc to the object. The content of the added data is an xml-construction which contains a
corresponding quantitative clarification. For example in an ontology for spatial data the
inclusion of a given object in another object is presented with Arc-object[Classification] and
quantitative clarification representing the transformation of the coordinate system of the
pointed (with the arc) object to the coordinate system of the pointing (with the same arc) object.
For a text presentation of the two models of arcs and its specialization we use the following
notation:
A. For Arc-object[Classification] – Clsf:<Name 1>, < Name 2>;
B. For Arc-object[Feature] – Ftr:< Name 3>, < Name 4>;
Where:
• „Name 1“ and "Name 3" are the names of the objects, the identifiers of which is entered
as the content of the Clarifying concept [Value] in the arcs
• „Name 2“ and "Name 4" are the names of the Objects[Segment], whose xml-
representation is entered as the content of the Clarifying Data[Value]
• If the specialization is only conceptual, then both representations have a short form:
Clsf: <Name 1>; and Ftr: <Name 3>;
19. uKG-construction Role.co
With the arcs of the uKG is presented the semantics of connectivity only through two arc
models and its specialization. This is not enough to present all the variants of semantic
connectivity. For example, in order to present a father-son relationship between two
individuals, in addition to a classification meaning reflecting the relevant type of semantic
construction and the specialization "Kinship" of the rainbow, it is necessary to indicate the type
of relationship. Practice shows that in many cases the additional semantics with which an arc
Unified Knowledge Graph Page 5
Inato Ltd, Sofia, 2022
must be loaded in uKG is significantly more, and sometimes with a much more complex
construction of its representation.
Therefore, in order to preserve the unification in uKG, such additional semantics is transferred
to an intermediate object, through which the corresponding connection is realized and thus the
unification of uKG is not violated. This intermediate object expresses some role in which the
object appears. For example, when reflecting the kinship connection "John is a father of Jim",
the additional semantics in the arc from John to Jim is "is a father of„ ".
This semantics can be transferred to an intermediate object - Jim [Father] of the class
"Role"(Role.co)
2
, which is with a model "Father". All objects from Role.co point to an object
that is a reference object for them - they refer to him the semantics contained in such class of
objects.
So the uKG-unification of the expression "John is a father of Jim" can be noted as follows:
Jim[Human] Clsf:Kinship; John[Father] Clsf: Kinship; John[Human].
The invers expression “Jim is a son of John” can be noted as follows:
Jhon[Human] Clsf: Kinship; Jim[Son] Clsf: Kinship; Jim[Human]
The intermediate objects (John[Father] and Jim[Son]) are Role.co. They are created by using
the Father[Model] and Son[Model].
Questions instead of conclusions
The unification of the KG created a prerequisites for semantically controllable navigation in the
graph topology. But every navigation path has a beginning and current end. In UPI there is a one
beginning node for Sysadmin which ensures the access to the whole uKG for the purpose of
technical servicing. Every (human) user of our UPI-based systems has their beginning node in the
graph topology. The same goes for the embedded in the UPI-environment bot-configurations,
because they have the same as the user’s identification in the UPI-environment. From semantical
point of view the UPI-users and the UPI-bots are indistinguishable and will be called “agents”
further in this article
Talking about uKG, it can be said that every agent has access to a part of the knowledge in the
uKG. The amount of this knowledge represents the cognition of that agent. The collaboration
between agents, as is presented in [6] needs a common understanding about the physical
environment of collaboration and its organization. This "common understanding" can be accepted
as a term for commonsense ( [7], [8], [9] and others), but this term does not have a well-defined
meaning nor scope of implementation.
That is why we have focused our attention on different semantic sets, potentially usable by more
than one agent, without adhering to any vision of common sense. Analyzing them, we used the
main feature of SNBA
3
, the unification of the KG and the introduced by us specific constructions
of Roles of a Reference objects and Classes of objects. This allowed us to discover for things in
the real world a lot of semantic sets with analogous constructions, but with different semantics.
This led us to a new understanding of the widely used term "ontology". We will share our
developments on this topic in a separate article.
The introduction of the unification of the KG allows us to look for another solution to the process
of knowledge input in the uKG. The UPI user interface is not convenient for large amounts of
descriptions. Nevertheless the biggest disadvantage is the difficulty of easily explaining the
introduced knowledge.
The biggest drawback lies in the process itself of introducing knowledge. The source form of
knowledge representation is a text in natural language. That representation is translating to the
language of the UPI user interface language. With the same language can be read the available
2
This can be presented in short as "Jim[Father] is from Role.co".
3
Let point again that this is the coexistence and use in a common environment of the models of concepts, data and
processes and their instances.
Unified Knowledge Graph Page 6
Inato Ltd, Sofia, 2022
knowledge in the uKG. But the understanding of descriptions in the "UPI-language" needs a
translation into natural language. The two translations- from natural language to the "UPI-
language" and back are one of the essential problems. Another essential problem is the poor
readability of knowledge represented with the "UPI-language", which means a poor explainability.
This had led us to the necessity of creating a very simplified, but fit for formalization version of
the Natural Language. We will share our developments on this topic in a separate article.
Bibliography
[1]
Lyubo Blagoev, Tihomir Blagoev, "Semantic Network Based Architecture," ResearchGate,
2019.
[2]
Lyubo Blagoev, Kamen Spassov, "Semantics representation regarding establishing and
maintaining a Semantic Interoperability in the e-Governance's environment,"
ResearchGate, 2020.
[3]
Cayley Wetzig, Knowledge Graphs, Property Graphs and TigerGraph, MEDIUM, 2022.
[4]
Michelle Knight, What Is a Property Graph?, DATAVERSITY, 2021.
[5]
Thomas Frisendal, Property Graphs Explained, Graph Data Modeling, 2022.
[6]
Lyubo Blagoev, Kamen Spassov, "Smart Home as a Digital Environment," Sofia University,
Spring Scientific Session of Faculty of Mathematics and Informatics, 2015.
[7]
David Gunning, Machine Common Sense Concept Paper, DARPA/I2O, 2018.
[8]
Keith Darlington, Common Sense Knowledge, Crucial for the Success of AI Systems,
OpenMind BBVA, 2020.
[9]
Walid S. Saba, Commonsense Knowledge, Ontology and Ordinary Language, Washington,
DC: Int. J. Reasoning-based Intelligent Systems,, 2008.
[10]
Achille C. Varzi, Words and Objects, Dordrecht/Boston/London: Kluwer Academic
Publishers, pp. 49–75, 2002.
[11]
Sam Kenyon, Mechanisms of Meaning, Medium, 2021.
