This paper is an extended version of a contribution presented
at the Graphiñon 2025 conference.
The
modern digital space is characterized by exponential growth in information
volumes and a qualitative increase in the complexity of information flows.
According to various estimates, over two and a half quintillion bytes of data
are generated daily in the global information space [1], a significant portion
of which represents unstructured information in the form of text content,
multimedia materials, and metadata structures. Given this information
abundance, researchers and practitioners are faced with the fundamental
question of developing adequate methods for assessing the information
completeness and qualitative characteristics of data available for real-world
objects of interest.
Traditional
approaches to the quantitative assessment of information resources, based on
simple metrics of data volume, frequency of mentions, or statistical indicators
of occurrence, demonstrate significant limitations when addressing the complex
characterization of the information value of aggregated data. The existing
arsenal of bibliometric and scientometric analysis methods [2], despite the developed
mathematical apparatus for assessing the scientific impact of publications and
citations, does not provide the tools for a comprehensive assessment of the
information environment of arbitrary objects beyond the scientific and academic
sphere. Similarly, the methodological arsenal of social network analysis,
focused on assessing the popularity and influence of nodes in network
structures, does not take into account the qualitative parameters of
information content and does not address the problem of assessing the
uniqueness of information units.
This
issue is particularly pressing in the context of the rapid development of
artificial intelligence systems and machine learning technologies. The
performance of modern neural network architectures is determined by the
qualitative characteristics of training data, including not only its
quantitative completeness but also its informational reliability,
representativeness, and absence of redundancy. However, currently dominant
approaches to training dataset formation focus primarily on quantitative
aspects of sample populations, such as sample size and the balanced
representation of different object classes, neglecting the qualitative
assessment of the information richness of the data relative to the entities under
study.
Similar
methodological challenges arise in related areas of information technology,
including information retrieval and knowledge extraction systems, recommender
systems, automatic natural language processing technologies, and the
construction of semantic knowledge graphs. These systems, whose functional
purpose is to aggregate, structure, and intelligently analyze information about
real-world objects of various natures, including organizational structures,
events, and abstract concepts, require the development of tools for adequately
assessing the completeness and qualitative characteristics of available
information.
Research
works in the subject area of entity resolution and knowledge graph construction
develop methodological approaches to linking and aggregating heterogeneous
information about research objects [3-4], but focus mainly on the technical
aspects of integrating heterogeneous data, without offering solutions for the
problem of assessing the information density and qualitative characteristics of
the resulting object representations.
Scientific
research in the field of information retrieval and extraction has developed a
sophisticated mathematical apparatus for assessing the relevance of documents
to user queries and information needs, including classical metrics of search
accuracy and completeness. However, existing methods do not solve the
conceptually distinct problem of comprehensively assessing the information
environment of an object as an integral phenomenon of digital space.
This paper
proposes a conceptual model for analyzing information space based on the
concept of the density of an object's information field. An object's
information field is defined as the totality of all information units
containing references, descriptions, or links to the object under study and
available in open digital sources. A fundamental characteristic of such an
information field is its density, which is an integral measure reflecting not
only the quantitative parameters of available information but also its
essential qualitative characteristics, including the uniqueness of the
information content, its relevance to the object of study, the authority of
information sources, and the temporal relevance of the presented data.
The
primary objective of this study is to develop the theoretical foundations of
the concept of information field density and formulate a methodological
framework for its practical application in the analysis and comprehensive
assessment of digital information resources. Achieving this goal requires
addressing a set of interrelated research tasks, including formulating rigorous
definitions of the basic concepts of an object's information field, information
field density, and information quantum, and developing a mathematical model for
calculating information field density, taking into account the multidimensional
characteristics of information content.
Modern
scientific literature contains a wide range of studies devoted to various
aspects of the assessment and analysis of information resources; however,
existing approaches demonstrate fragmentation and limited applicability to the
task of comprehensively assessing the information environment of arbitrary
real-world objects.
The
fundamental theoretical foundations of quantitative information assessment were
laid in Claude Shannon's classic works on mathematical information theory,
where the information content of a message is determined through the entropic
characteristics of the source and transmission channel. Shannon's concept of
information entropy [5] provides a mathematical framework for measuring the
uncertainty and information capacity of systems. However, this approach
operates exclusively with the syntactic characteristics of information,
ignoring the semantic content of messages and their pragmatic value for solving
specific problems. The development of this approach in Andrey Kolmogorov's
works on algorithmic information theory [6] introduces the concept of string
complexity as the length of the shortest program capable of reproducing it,
which allows us to formalize intuitive notions about the meaningfulness and
non-randomness of information sequences.
A
significant body of modern research is focused on the problem of assessing data
quality [7], considered as a multidimensional characteristic of information
resources. Conceptual models of data quality, developed in the works of
researchers in this area, include such fundamental dimensions as the accuracy of
information content, reflecting the correspondence of data to the actual state
of the described objects, the completeness of representation, characterizing
the degree of coverage of relevant aspects of the subject area, data
consistency, determining the absence of internal contradictions in information
structures, the relevance of information relative to the temporal context of
their use, and the relevance of information in relation to the assigned
analytical tasks. These studies propose a variety of metrics and algorithmic
approaches to the automatic assessment of the qualitative characteristics of
data, including statistical methods for identifying anomalies and
inconsistencies, techniques for comparative analysis of multiple sources to
verify actual accuracy, and heuristic algorithms for assessing completeness
based on the structural features of subject areas. However, existing data
quality studies demonstrate a methodological focus primarily on structured
information resources, such as relational databases and formalized catalogs,
without offering adequate solutions for assessing the qualitative
characteristics of unstructured information that naturally aggregates around
objects in open digital space.
Entity
resolution and semantic knowledge graph construction methods aimed at
integrating and structuring heterogeneous information about real-world objects
have attracted significant research attention in recent decades. The task of
entity resolution, which involves identifying and linking different mentions of
a single object across heterogeneous information sources, is solved through a
combination of string representation comparison methods, contextual analysis of
mentions, and machine learning on labeled data corpora. Modern approaches to
knowledge graph construction, such as those implemented in Google projects,
Knowledge Graph, DBpedia, and Wikidata [8-10] demonstrate impressive results
in aggregating structured information about millions of entities of various
types, providing unified interfaces for accessing integrated knowledge.
However, existing research in this area focuses primarily on the technical
aspects of integrating and verifying structured data, without offering
methodological solutions for assessing the information density and qualitative characteristics
of unstructured information environments of objects, which cannot be adequately
represented in the format of structured graph relations.
Classical
research in information retrieval and extraction has developed a sophisticated
mathematical framework for evaluating the effectiveness of search engines and
the relevance of documents to user queries. Fundamental metrics of precision
and recall, as well as their harmonic mean in the form of the F-measure,
provide quantitative tools for assessing the performance of information
retrieval systems under controlled experimental conditions. Developments in
this field have led to the creation of complex relevance models that take into
account multiple factors, including the textual similarity between the query
and the document, the authority of information sources, the temporal
characteristics of documents, user behavior, and the contextual features of
information needs. Modern approaches to document ranking in web search, based
on algorithms such as PageRank [11] and its modifications, demonstrate the
effectiveness of taking into account structural characteristics of the
information space, such as hyperlink and citation patterns, to assess the
authority and significance of information resources. However, existing methods
of information retrieval solve a conceptually different problem of assessing
the compliance of individual documents with specific information requests,
without offering approaches to a comprehensive assessment of the information
environment of objects as holistic phenomena of digital space, characterized by
specific patterns of information aggregation and qualitative diversity of
sources.
Bibliometric
and scientometric studies have developed sophisticated methods for assessing
the scientific impact and significance of publications, based on the analysis
of citation patterns and collaborative relationships between researchers.
Classic indicators, such as the h- index and its many modifications, provide
comprehensive assessments of the productivity and influence of scientists,
taking into account both quantitative characteristics of publication activity
and qualitative parameters reflected in the frequency of citations of works by
the scientific community. Modern approaches to scientometric analysis include
network methods for studying scientific collaborations, temporal analysis of
the evolution of research areas, and interdisciplinary metrics reflecting the
broad impact of scientific results. Despite the methodological sophistication
of bibliometric approaches, their applicability is limited to the specific
context of scientific and academic activity and does not extend to assessing
the information characteristics of arbitrary objects unrelated to formal
scientific citation systems and peer review.
Research
in the field of the Semantic Web and ontology engineering offers formalized
approaches to representing and structuring knowledge about subject areas, based
on logical formalisms and standardized resource description languages.
Ontological models provide expressive means for specifying the conceptual
structures of subject areas, including hierarchies of object classes,
properties and relationships between entities, integrity constraints, and
inference rules [12–13]. Linked data technologies developed within the
framework of the Semantic Web initiative demonstrate the practical
effectiveness of integrating distributed information resources through
standardized protocols and data representation formats. Methods for assessing
the quality of ontologies include analysis of the logical consistency of
conceptual models, assessment of the completeness of subject area coverage, and
metrics of the correspondence of ontological structures to the real
characteristics of the phenomena being described. At the same time, ontological
approaches presuppose the presence of expertly developed conceptual models and
do not solve the problem of assessing the qualitative characteristics of
spontaneously formed unstructured information, which cannot be adequately
described within the framework of pre-defined ontological schemes.
An
analysis of existing research areas reveals a fundamental gap in the
methodological arsenal of modern information technologies, stemming from the
absence of conceptual approaches to the comprehensive assessment of information
fields naturally forming around arbitrary objects in unstructured digital
space. Existing methods, despite their development within specialized subject
areas, do not provide adequate tools for the integrated assessment of the
qualitative characteristics of heterogeneous information aggregated around
objects of various natures in the information abundance of the modern digital
space. This methodological vacuum necessitates the development of new
conceptual approaches capable of providing a theoretical foundation for the
comprehensive assessment of information density and the qualitative
characteristics of object-oriented information fields.
Developing
a conceptual framework for analyzing the information fields of objects requires
the formulation of new theoretical constructs that would overcome the
limitations of existing approaches to assessing information resources. The
proposed theoretical model is based on a synthesis of classical information
theory, modern concepts of semantic space, and the principles of systems
analysis of complex information structures. The central premise of this
approach is the understanding that information about real-world objects does
not exist in digital space as isolated discrete units, but rather forms complex
interconnected structures that can be conceptualized as information fields with
specific topological and qualitative characteristics.
The
concept of an object's information field is based on the metaphorical transfer
of physical concepts of field structures to the analysis of information
phenomena. Just as physical fields are characterized by the distribution of
energy or matter in space, an object's information field represents the
distribution of information units in digital space, where each information unit
possesses a certain "mass" or significance, and the aggregate of such
units forms a complex topological structure with various zones of information
concentration and sparseness. This conceptualization allows for the application
to the analysis of information structures of the mathematical apparatus
developed for the study of field phenomena, including the concepts of density,
gradient, flow, and other characteristics describing the spatial distribution
of physical quantities.
A
formal definition of the information field of an object can be formulated as
follows: the information field of an object O is the set of all information
units I = {i₁, i₂, ..., iₙ} available in the digital space
and containing direct or indirect references, descriptions, links, or any other
forms of informational connection with this object. Mathematically, this can be
expressed as IF(O) = {i
∈
I | R(i, O) >
θ}, where R(i, O) represents the relevance function of the information
unit i with respect to the object O, and θ is the threshold value
determining the minimum degree of relevance for including the information unit
in the object's field. It is important to note that the boundaries of the information
field are blurry, since the relevance of information can vary widely, from
direct references to an object to complex contextual associations, the
establishment of which requires deep semantic analysis.
An
object's information field is characterized by a number of fundamental
properties that determine its structure and dynamic characteristics. Spatial
heterogeneity of the field manifests itself in the fact that different areas of
digital space contain an uneven distribution of information about the object.
Some sources and platforms may accumulate significant volumes of relevant
information, while other segments of the information space contain virtually no
mention of the object. Temporal dynamism is another key property of information
fields, as they continuously evolve under the influence of new information
sources, updates to existing materials, changes in the object's popularity, and
other factors influencing information activity around it. The
multidimensionality of the information field is due to the fact that
information about an object can exist in various formats and at various
semantic levels, including factual data, analytical materials, multimedia
content, meta-information, and contextual relationships.
The
central characteristic of an information field is its density, which represents
an integral measure of the concentration of qualitative information about an
object within a unit of information space. Unlike simple quantitative metrics
such as total data volume or frequency of mentions, information field density
must take into account the qualitative parameters of information content,
including the uniqueness of the information, its relevance to the object of
study, the authority of sources, and the temporal relevance of the data. Conceptually,
information field density can be understood as a measure of the information
"mass" of an object in digital space, reflecting not only the amount
of available information but also its ability to form a complete and accurate
representation of the object of study. More generally, information field
density characterizes the ability of a technology, such as artificial
intelligence, to recreate an image of an object based on collected data [14],
established patterns, and relationships.
It's
worth noting that this concept represents an attempt to move away from simply
quantifying information in bits or the number of tags, links, or mentions.
Given the existence of modern language models, information acquires particular
value when it can be interpreted and searched using not just words, but
semantic constructs and contexts. In terms of quantitative measurement of
information, the traditional assumption is that the more data available, the
more accurately and completely a digital image of an object can be recreated.
However, the heterogeneity of information concerning the same object of study,
across different interpretations, makes the idea of creating a neutral model
whose answer satisfies all parties virtually impossible. As the NewsGuard
report [15-16] states, the growth of information coverage and the connection of
web search to language models and chatbots has resulted in a noticeable
deterioration in the accuracy of the results returned, and during events taking
place “online”, chats more often reinforced false narratives, pulling in
materials from dubious sources and making no distinction between authoritative
publications and their propaganda lookalikes.
At
the same time, an integrated assessment of the density of information, both
true and false, surrounding an object can allow for an analysis of the volume
of sources a model can potentially rely on, as well as what image of the object
it is most likely to form, what patterns it will utilize, and what opinions it
will prioritize. Given that, in light of the development of language models
across political poles, the market will likely push AI services toward a more
explicit position in an attempt to meet the expectations of their audiences,
there is a need to create neutral and objective criteria for assessing
information volumes in order to interpret the main trends in the development of
digital images of certain phenomena that language models will focus on. In the
near future, the development of such services may lead to different language
models, limited by fundamentally different principles, providing diametrically
opposed answers to the same question. In light of this, assembling a relatively
objective information picture will become increasingly difficult with each
passing year.
The
density of an object's information field can be characterized by the number of
conventional units (a conventional unit is a quantum of unique, word-for-word,
non-repeating information containing a
thought/analysis/research/fact/conclusion related to the object in question). Then,
say, 1,000 generated thoughts related to the object will increase the density
of its information field, while, for example, 1,000 reposts (1,000 mentions of
the same term without the slightest change in wording) will not change the
density level. In this case, the quantity of information in the traditional
sense of computer science, i.e., measured in bits, may be only one parameter,
but not the most significant one.
In
this regard, we can introduce a name for such a conventional unit. Infon (from
"information" + the suffix "on") – in the primary
definition, these are non-repeating units of information containing original
thoughts, analysis, research, or facts about an object.
However,
each conventional unit (quantum of information) must, in one way or another,
have its own "weight" or "significance." For example, a
scientific article with original research on a subject and a random comment on
a social network, although both contain unique information, have different
values for shaping the information field. Thus, "heavier" information
units have a stronger influence on the information field. It is also important
to consider the time component. Information tends to become outdated, and its
significance can change over time. For example, a ten-year-old scientific
article may have less weight than a more recent study, unless it is fundamental
to the field.
An
interesting addition could be the concept of "information
resonances." When multiple independent sources confirm the same
information in different ways (not simply copying), this can create a
reinforcing effect—similar to how waves can reinforce each other during
resonance. Such resonances can significantly and sharply increase the density
of the information field at certain points in time. Visualizing such temporal
fluctuations can significantly improve understanding of the development of a
given phenomenon.
Another
important aspect is the coherence of such information quanta. Individual
fragments of information, linked by logical or cause-and-effect relationships,
can form more stable and meaningful structures in the information field than
isolated facts.
If we
consider the infon from the perspective of these aspects, its definition
becomes more precise: an infon is the minimal indivisible unit of unique
information about an object, which cannot be reduced without loss of semantic
content. An infon is the minimal unique unit of information containing a
complete thought (a fact) about an object, which is not a direct repetition of
existing information. This model is open to two interpretations: one can view
infons as a unit of measurement, in which case they must be identical. This
necessitates defining what constitutes the minimal unit of unique information.
Another
approach to the model assumes that each infon is defined as a quantum of unique
information and can have its own informational weight and relevance (Fig. 1).
An infon is characterized by atomicity, meaning the impossibility of its
further division without loss of semantic and informational integrity;
uniqueness, implying the absence of exact duplicates of a given information
unit in other sources; relevance, ensuring a direct connection between the
content of the infon and the object of study; and verifiability, allowing for
the verification of the factual accuracy of the information contained in the
infon .
Figure
1 shows an example of a possible visualization of such a phenomenon, where the
represented object is surrounded by heterogeneous particles graded in color and
size. The size of each particle reflects the size of the information quantum,
while the color demonstrates the degree of conceptual connection with the
object, from directly related to the object to very distantly related.
Fig. 1. Example
of visualization of the density of the information field of an object
The
process of identifying information from a general information field is a
complex analytical task that requires a combination of automatic natural
language processing methods and expert content analysis. Algorithmic
identification can be based on text segmentation methods to identify
semantically complete fragments, information novelty analysis through
comparison with existing knowledge bases, assessment of actual uniqueness
through plagiarism and duplication detection, and determination of relevance
using semantic analysis and machine learning methods. Each identified
information can be characterized by a set of quantitative parameters, including
its uniqueness, its degree of relevance to the object, an assessment of the
reliability of the information contained, and an indicator of information value
for forming a holistic understanding of the object.
Since
information in the modern world is presented in a wide variety of forms
(visual, audio, video streams, text, social media comments, and numerical data
sets), combining them into a unified structure becomes a distinct challenge.
The key to unifying different types of information lies in the concept of a
multidimensional information space. An information field can be imagined as a
multilayered structure, where each type of data forms its own layer, but all
are interconnected and influence each other. This is similar to how different
modalities (text, images, sound) can be transformed into a single vector space
in neural networks.
The concept of “information embeddings” is suitable for unifying different types of data – the transformation
of any type of information into a universal vector representation [17]. Modern
technologies already make this possible: CLIP can find connections between text
and images, wav2vec converts audio into vectors, and large language models
transform text into multidimensional representations. In this case, it is
important to consider the "information density coefficient" for
different types of data. For example, one second of video may convey more
information about an object than a text description of the same duration, but a
textual analytical article may contain deeper semantic information than a
simple photograph.
The question of "cross-validation" between different types of data also arises.
Validation between different types of data. If information from
different sources and formats corroborates each other, this increases the
credibility of each individual piece of information. For example, if a textual
description of an event is supported by video footage and numerical data, the
overall credibility of the information increases. In the context of artificial
intelligence, this approach opens up new possibilities for creating multimodal
systems capable of forming a holistic view of objects based on heterogeneous
data. This could prove to be another data organization system useful for the
development of general artificial intelligence systems, which must be able to
process information holistically, similar to the human brain.
Currently,
large generative artificial intelligence models tend to accumulate numerous
individual algorithmic rules—specialized patterns for specific cases—that do
not integrate into a coherent knowledge system. Such localized patterns often
contradict each other, creating internal conflicts in the system's operation.
Research attempts to find coherent conceptual representations in the model
structure [18] reveal only disparate information fragments that do not form a
single, coherent image. Nevertheless, such distributed rules have a certain
practical value. The enormous parametric capabilities of language models allow
for the storage of such patterns in large quantities, and the quantity often
compensates for the lack of a clear structure. The ability to create verifiable
internal representations opens avenues for combating artificial hallucinations,
increasing the reliability of logical inferences, and ensuring greater
transparency in the operation of intelligent systems.
The
mathematical formalization of the density of the information field can be
represented as a weighted sum of the information contributions of individual
elements of the field, where each element (infon) is assessed according to
multiple quality criteria. Let the information field (IF – Information The
Field object is a multidimensional vector space. The basic formula might look
like this:
where Qi is a certain information quantum,
Wi is a weighting coefficient reflecting
the type and significance of the information unit iᵢ,
T(iᵢ) takes into account the temporal relevance of the data,
R(iᵢ) determines
the relevance relative to the object, K is a certain normalizing coefficient.
The formula can also be supplemented with parameters such as
A (iᵢ), which
reflects the authority of the information source, and U(iᵢ), which
characterizes the uniqueness of the information content.
Each
of these coefficients has its own calculation methodology based on objective
parameters and existing data analysis methods. It makes sense to seek out and
borrow some of these coefficients from existing research on big data . There
are a number of related research areas: information field theory in physics
[19], semantic spaces in linguistics [20 link to Klyshinsky /elastic maps],
digital information ecology, and quantum information theory. These areas
provide useful tools and methodologies that can be adapted to develop
information field theory.
For
example, if we consider Wi (the source weighting factor), the closest analogs
would be entities such as the impact factor of scientific journals, Google 's
PageRank, and the citation index. A possible calculation formula might be
something like this:
where As (Authority
Score) is the authority of the source (0-1),
Cs (Citation Score) is
responsible for the citation index, Rs (Reliability Score) characterizes the
reliability indicator based on historical data,
Vs (Verification Score) is the possibility of verifying information, and
Nmax is a certain normalizing maximum
Ti metric
can also be used in real-world examples: scientific databases consider the
"age" of publications when calculating their importance, and Netflix
uses similar time-based factors to rank content. Metrics for assessing
information relevance (Ri) are used in modern search engines and
natural language processing (NLP) systems.
The
normalizing coefficient is determined empirically for a specific subject area
and can be calculated as:
where max (IF) is the
maximum possible value of the information field in a given area.
The
interaction between the concepts of the information field, its density, and its
constituent infons forms a theoretical model that enables the qualitative
analysis of information structures in digital space. Infons act as individual
particles of the information field; their totality defines the field's
structural characteristics, while their qualitative parameters determine the
overall density of an object's information field. This model provides a
theoretical foundation for developing practical methods for assessing
information resources, enabling a transition from intuitive notions of the
"richness" or "poverty" of an object's information to more
rigorous quantitative assessments based on an analysis of the qualitative
characteristics of its information content.
The
proposed approach to assessing information density in the digital environment
opens up new opportunities for practical application and further development.
Key areas of potential application and development prospects for the
methodology include information security, social media and content analysis,
and data preparation for neural network training.
In
the field of information security, the methodology for assessing information
density can be applied in several key areas. First, analyzing the density of
the information field can help identify abnormal spikes in activity that may
indicate targeted information campaigns or attacks. Second, assessing the
qualitative characteristics of the information space will ultimately help
identify sources of unreliable information and monitor the spread of
disinformation. Early detection of information threats by analyzing the
dynamics of changes in information density may also prove relevant.
In
the context of social media analysis, the proposed methodology can provide
tools for a deeper understanding of information processes. Assessing
information density allows for identifying significant trends and separating
them from information noise, which is especially important in the context of
content overload on social media. Analyzing the qualitative characteristics of
the information field helps determine the real impact of content and its
authors, going beyond simple quantitative metrics such as the number of likes
or shares .
In
light of the development of this theory, the following directions for further
research can be identified.
1. The
development of mathematical apparatus is one of the key areas. It is necessary
to develop more accurate models to describe the interactions of various
components of the information field and to create methods for quantitatively
assessing the qualitative characteristics of information. Particular attention
should be paid to the creation of mathematical models that take into account
the temporal dynamics of information processes and the nonlinear nature of the
interactions between different types of information.
2. Experimental
verification of the methodology requires conducting a series of studies in
various subject areas. The effectiveness of the proposed methods for assessing
information density must be confirmed using real data, and the results must be
validated in various application contexts. An important aspect is the development
of standardized experimental methods and criteria for evaluating the results.
3. The
creation of practical tools is a necessary step for the widespread
implementation of the methodology. This requires the development of software
capable of automating information density analysis processes, the creation of
user-friendly interfaces for working with data, and integration with existing
information analysis systems.
The
development of the proposed methodology could have a significant impact on
several aspects of information technology. In the field of search engines, it
will enable the creation of more accurate search ranking algorithms that take
into account not only quantitative but also qualitative characteristics of
information. In the field of artificial intelligence, the methodology could
facilitate the development of more sophisticated natural language processing
and data analysis systems. This methodology could also contribute to the
development of personalized recommendation systems that can more accurately
account for context and information quality.
This
study is conceptual in nature and aims to formulate the theoretical foundations
of a new approach to analyzing information space through the lens of the
concept of information field density. The main result of this work is the
introduction of a system of interconnected concepts, including an object's
information field, its density, and its constituent information quanta (infons),
which together form a holistic conceptual model for the qualitative
assessment of information resources in digital space.
The
mathematical formalizations proposed in this study are primarily illustrative
in nature and serve to demonstrate the fundamental feasibility of
quantitatively describing the qualitative characteristics of information
structures. Further development of a rigorous mathematical framework will
require extensive empirical research to determine the specific parameters of
the uniqueness, relevance, authority, and other components of the proposed
density model. Particular attention should be paid to the operationalization of
the concept of infon, which requires the development of algorithmic procedures
for automatically extracting information quanta from unstructured text arrays
and qualitatively assessing them.
The
theoretical significance of the proposed approach lies in its ability to
overcome the limitations of existing methods for assessing information
resources, which traditionally focus either on quantitative data characteristics
or on highly specialized aspects of information quality. The practical
significance of the developed concept lies in its potential applications in the
development of next-generation artificial intelligence systems capable of
generating more reliable and verifiable internal representations of real-world
objects. The transition from statistical patterns extracted from uncontrolled
text corpora to a systematic analysis of information density can significantly
reduce the frequency of generating unreliable information in language models
and increase the transparency of decision-making processes in intelligent
systems. The presented conceptual model can thus serve as a starting point for
the development of a new research paradigm in the field of information resource
analysis and their qualitative characteristics.
1. Austin Harris, 2.5 quintillion bytes of data are produced by people every day, 2021, - URL: https://appdevelopermagazine.com/2.5-quintillion-bytes-of-data-are-produced-by-people-every-day/
2. Malakhov V.A., Bibliometric analysis as a method of science studies: possibilities and limitations // Science studies. 2022. No. 1. URL: https://cyberleninka.ru/article/n/bibliometricheskiy-analiz-kak-metod-naukovedcheskih-issledovaniy-vozmozhnosti-i-ogranicheniya (date of access: 09/08/2025).
3. Hogan A. et al. Knowledge graphs //ACM Computing Surveys ( Csur ). – 2021. – T. 54. – No. 4. – S. 1-37.
4. Kislitsyna M. Yu . Analysis of the Error Structure in Identifying the Author of a Text Using the Nearest Neighbor Graphs (2025). Scientific Visualization 17.2: 110–122, DOI: 10.26583/sv.17.2.08
5. Shannon C.E. A Mathematical Theory of Communication (1948)
6. Kolmogorov, A. N. (1965). "Three approaches to the quantitative definition of information"
7. Batini C. et al. Data and information quality //Cham, Switzerland: Springer International Publishing. – 2016. – T. 63.
8. Singhal A. Introducing the Knowledge Graph: things, not strings (2012) URL: https://blog.google/products/search/introducing-knowledge-graph-things-not/
9. Dbpedia : Global and Unified Access to Knowledge Graphs, URL: https://www.dbpedia.org/
10. Wikidata, URL: https://www.wikidata.org/wiki/Wikidata:Main_Page
11. Brin S., Page L. The anatomy of a large-scale hypertextual web search engine //Computer networks and ISDN systems. – 1998. – T. 30. – No. 1-7. - WITH . 107-117.
12. SI Chuprina, IA Labutin . A High-Level Adaptation Toolkit for Unified Integration of Software Systems with Neural Interfaces (2024). Scientific Visualization 16.4: 11–24, DOI: 10.26583/sv.16.4.02
13. SI Chuprina . Using Data Fabric Architecture to Create Personalized Visual Analytics Systems in the Field of Digital Medicine (2023). Scientific Visualization 15.5: 50–63, DOI: 10.26583/sv.15.5.05
14. NA Bondareva . The Impact of Input Data Density on the Performance of Graphic Neural Networks (2024). Scientific Visualization 16.5: 109 - 119, DOI: 10.26583/sv.16.5.08
15. Morrone M., Exclusive: Popular chatbots amplify misinformation, 2025, - URL: https://www.axios.com/2025/09/04/popular-chatbots-amplify-misinformation?utm_source=Securitylab.ru
16. AI False Information Rate Nearly Doubles in One Year, 2025, - URL: https://www.newsguardtech.com/ai-monitor/august-2025-ai-false-claim-monitor/
17. Li S., Guo H., Tang X., Tang R., Hou L., Li R., Zhang R. Embedding Compression in Recommender Systems: A Survey (2024), arXiv:2408.02304
18. Vafa K., Chen JY, Rambachan A., Kleinberg J., Mullainathan S. Evaluating the World Model Implicit in a Generative Model (2024) https://doi.org/10.48550/arXiv.2406.03689
19. Torsten A. En?lin . Information theory for fields (2018). Annalen der Physik 2019, vol. 531, issue 3, p. 1800127 DOI: 10.1002/andp.201800127
20. Bondarev A.E., Bondarenko A.V., Galaktionov V.A., Klyshinsky E.K. Visual analysis of cluster structures in multidimensional volumes of text information / Scientific visualization, Vol. 8, No. 3, 2016, pp. 1-24
