Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Semantic Representations of Political Discourse:

based on Topic Modeling and Algorithms
DAGIIMAA BALAANZ
School of language and culture,
University of the Humanities Ulaanbaatar, Mongolia

Abstract:- This research is a contribution to text expresses in words what he wants to say. In other words,
network analysis using text mining algorithms, how the when using the function of language to name a
speaker creates discursive approaches of phenomenon and object, the speaker begins by describing
macrostructure and interfaces with knowledge based on what has been being said.
the mental model within discourse macrostructure
theory, developed by the Fairclough, Teun A. van Dijk, Causal function/uchir shaltgaant/ - In theory, the
aimed to identify and analyze authors opinion and causal function is a means of expressing ideas (Finch 1998,
attitude in political speech. The detection of sentiment Wisniewski, 2007). Because the human brain processes
and emotional tone is also related to the method of information, it is very difficult to think without words.
semantic structure in discourse, which is a psychological When a person thinks about something or a concept, he
factor. Therefore, the episodic and mental space, which puts it into words in his mind. For example, if someone
study of the psychological factors based on the theory of thinks about stationery, he thinks of pencils, books, lines,
knowledge, has practical importance in determining erasers, and pencils. which is expressed in words.
the thought. In the social and language sciences,
knowledge is also studied to measure variables and Function of Communication /khariltsaanii uureg/ -
psychological factors for thought and its This means that most speakers consider language to be a
key communication tool. Wanting, begging, informing,
interrelationships.
commanding, promising, and resisting are all motivated by
Keywords:- Discourse macrostructure, text network mutual understanding. (Wisniewski, 2007)
analysis, discourse bias, mental models, topic modelling,
text mining, mathematic stats. Considering the role of language in the
macrostructure, it is responsible for expressing the logic of
I. INTRODUCTION ideas. These include function of Imagination: Theoretically,
it is the process of creating awareness in mental activity,
The process of discussing an issue, expressing one's and the principles of understanding and imagining what is
views and ideas, and persuading others is a mental act of happening around people through language. (Finch 1998,
creating knowledge and expressing knowledge, and it is a Wisniewski, 2007). Textual function is used as a specific
manifestation of language use. Therefore, modern language tool and as a template for expressing ideas.
researchers study ideas in many ways, based on discourse
theory and semantics. Since the study of discourse theory Metafunction: Theoretically, metadiscourse uses the
and discursive approaches based on linguistic and role of language to express discursive approach and autor’s
psychological factors is interesting and modern, researchers attitudes play a role in the interaction. (Brown, Yule, 1983).
are trying to study the interrelationship between statistical Interactive roles involve social and personal feelings
methods and language issues. So, this study aims to around discourse. (Leech, 1983, хууд. 3) (Wisniewski,
determine whether a person's thought and knowledge can 2007) The concept of creating and understanding discourse
be determined by mathematical optimization of any on the concept of understanding and expressing ideas. (Van
decision, according to the use of the language and the Dijk in Wodak & Chilton, 2005, p. 71) The primary goal of
semantic structure, and to determine the global meaning the discourse is to use appropriate knowledge to the
and tone of speech at the macrostructure of discourse. situation, and as a result the listener understands the
situation and achieves mutual results (Van Dijk in Wodak
II. LITERATURE REVIEW & Chilton, 2005, p. 72) The concept of creating and
understanding discourse is interrelated with the ideas and
A. Language use for communication opinions expressed in the context. (Van Dijk in Wodak &
Language is the tool for conveying one's thoughts about Chilton, 2005, p. 71)
the role of language. (Wisniewski, 2007) Accordingly, the
primary functions of language begin with an understanding Discourse explores a wide range of cognitive-
of the nature of speech at the level of communication psychological fields, such as relationship formation, text,
(Bloomfield, 1993) and Finch's (1998) theory (Wisniewski, semantcs, and key factors influencing the psychology of the
2007) considered at the macrostructure. parties involved. (Van Dijk in Wodak & Chilton, 2005, p.
72) From above perspectives, purposeful communication,
Function to describe/todorkhoiloh uureg/ - language speech strategies, and speech methods within a given topic
is used to describe things and actions in the world. can be understood as discourse.(Ibid, p. 5)
(Wisniewski, 2007) People perceive and understand things
in their minds from their perspectives, and then the speaker

IJISRT22MAY581 www.ijisrt.com 72
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. THEORY OF DISCOURSE STRUCTURE examples, and conclusions. Parts of the structure have the
role of discourse macrostructure. Most relationships are
Discourse is a way of representing and interpreting nonlinear and binary. The most obvious part of a
ideas. The idea of the study of speech and text originated in macrostructure that comes to mind after reading is the use
the 19th century and the first half of the twentieth century of terms such as global meaning, macrostructure, and
(В.Фон Хумболдт, А.А.Потебня , Ф.Боас , В.Я.Пропп , macro propositions in discourse studies. In other words,
В.Матессиус, С.Балли), and different approaches to "memory for the text" is the general concept of meaning
speech and text The idea of studying (or linguistic analysis that remains in the mind after reading. The “global”
of text) was developed in the middle of the twentieth meaning of the discourse is explored by modeling the
century (IA Figurovsky, Z. Harris, K. Pike, P. Hartmann general idea through “topics” and “just”. It is a hierarchical
and others). From the 1970s to the 1980s, speech analysis structure of meaning that creates the logical sequence of
became as important as linguistic research. (В. Дресслер, text and speech unity and content. The rhetoric structure is
Т. ван Дайк, Э. Шеглофф, Р. Лонгакр, Т. Гивон, В. based on the basic principles of integrating the text, the
Чафе) Maintaining the Integrity of the Specifications. logical connection of meaning, and the relevance of the
topic.
IV. THEORY AND METHODS OF MODELING
KNOWLEDGE Integrating text: semantic and grammatical rules,
tools for connecting parts of the text, and their use. Logical
Knowledge is one of the main issues of discourse and relation of meaning: coherence, the logical sequence in
includes general and specific, semantic and episodic forms structure, optimal wording style, and the correct choice of
of knowledge of the parties involved in the relationship. vocabulary will be considered psychological factors in
Knowledge is an core factor in creating discourse and discourse.
understanding discourse which is the general
“understanding”level of a writer, speaker, or interlocutor, (Reinhart (1980 оны Yuan Wang & Minghe Guo,
how people discuss the issue from many views, how well it 2014, pp. 460-465 ишлэл). Organizing any information is
meets the general basic principles, how evaluates the a concept of the mind, cognition, and cognitive psychology,
relationship between the reall world and the phenomenon, and these actions are models for expressing the creation of
It is the timing of the issue between the times, and how knowledge. In the macrostructure of the text, the discursive
chronology is used, are the most important criteria for approach is a key factor of speech and even the smallest
determining the direction of an individual's attitude and component of speech creates it. Thus, the discursive
opinion. approach is a logical and understandable way of speaking,
and if the audience understands it well, the purpose of the
(Б.Дагиймаа) . The understanding is more important discourse will be achieved.
than creating discourse.(Нунань, 1993). The general
principles and concepts of knowledge creation are similar The general principles and concepts of knowledge
to the general concepts and principles of knowledge and are creation are similar to the general concepts and principles
defined by theory and models of cognitive psychology, of knowledge and are defined by models of cognitive
philosophy of knowledge, and general theory of mind. psychology, philosophy of knowledge, and the general
theory of mind. The knowledge based on cognitive
One of the ways in which knowledge is constructed psychology is mental models in the mind.
and analyzed is through the study of discourse and speech.
(Alvesson, Karreman, 2000a). The main concept of this A mental model is fact-based knowledge that arises
field is that language can be used more creatively than from a basic concept, outcome, or situation. The mental
patterns from simple to complex, a canonical model, and an model is a fundamental issue of language, behavior, and
approach based on certain constructivist epistemology is cognition, and is a complex set of thoughts and mindsets,
used in the study of discourse and speech. (Wood, Kroger, such as each individual's perceptions, perspectives, and
2000) Also, due to new trends in linguistics, new research worldviews. The mental model takes place in the human
styles are being incorporated into social change and new brain and is an expression of the person's inner knowledge
topics and styles are being created, which is a way to study of understanding the world.
discourse and speech. Exploring the discursive approach
and sentiment of discourse based on structure and patterns (Hall et al. 1994, Swan and Newell 1998, Sterman
of models has become a relatively new approach and 2000) There is a close relation between language and
presents many effective kinds of research. (Fineman, 1996; thought. In the political speech, ideology is discursive;
Mumby, Putnam, 1992). usually in the form of text or speech. (Dijk, 2006)
Therefore, it is the notion that the relationship, the content,
Text structure theory: Regarding the text structure and the discourse idea can be interpreted from different
theory, a theory that explains the logical connection and views, depending on the method used by the human and the
coherence of meanings in a text-based on the principle that computer.
each part of the text and speech has its special function,
there are no words or meanings, a hierarchical meaning Mental models, mental maps and mind maps are often
structure, and a unified introduction and conclusion. Text used to represent complex ideas and have a positive impact
Rhetorical structure is the sequence of words, reasons, on teaching and learning. Mental maps can be extracted
situations, concessions, arguments, explanatory notes, from texts to produce cognitive maps, which can then be

IJISRT22MAY581 www.ijisrt.com 73
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
used for sentiment analysis. In the recent years this field C. latent semantic analysis or LSA, pLSA algorithm
has been enhanced with the network metaphor: the nodes latent dirichlet allocation or LDA, relational topic
are the concepts and the connections are the relations models, word2vec algorithm, and its extension lda2vec.
between them The relations between the concepts can be These methods are based on retrieving the topics from text
plotted on the basis of their co-occurrences, which, in turn by identifying the clusters of co-occurrent words within
can facilitate the process of semantic priming — the them, based on the bag-of-words and skip-gram models.
process by which words tend to be recognized faster when This data can then be used to classify similar documents,
used with the words that have close semantic proximity to improve text indexing and retrieval methods, and to
them . Text network analysis methods based on this identify evolution of certain topics overs a period of time
approach have been shown to be successful for better within a specific text corpus.
comprehension of texts and topic modeling, (Paranyushkin,
2020). LDA provides the ability to get a better understanding
of the text network topology, which can yield many other
Topical modeling is a method of calculating the insights about the discourse structure in LDA or word2vec:
human mind and determining the group of ideas and words. how connected a discourse is, whether it’s biased towards a
Topical modeling is a key method used in text analysis to certain set of concepts, how dominant a specific topic is
determine keyword frequency. towards the whole discourse, whether there are structural
gaps that may indicate a potential for new ideas.
At the written and spoken level, the study of (Paranyushkin, 2020)
“information distribution and informative properties” by
mathematical methods of text analysis and text mining has D. Text-to-Network Conversion
become an important application in linguistics and The text is then converted into a directed network
communication in determining the mathematical meaning graph. The normalized words (lemmas) are the nodes in the
of words and making rational decisions in a short time. network graph and their co-occurrences are the edges.

V. ALGORITHS OF TOPIC MODELING TO a) Extracting Most Influential Keywords

REPRESENT MENTAL MODELS Using Betweenness Centrality, apply a ranking
algorithm to identify the nodes (the words) with the
A. IR Models highest betweenness centrality – these are the nodes
Text Normalization -Stop words Removal -Boolean that appear most often on the shortest paths between
Model – TF-IDF – Vector Model – Probabilistic Model – any two Two consecutive scans are performed. The
Latent Semantic Indexing Model – Neural Network Model first scan creates the connections (graph edges)
–Retrieval Evaluation – Retrieval Metrics – Precision and between the lemmas (nodes) that appear next to each
Recall – Reference Collection –Userbased Evaluation. other (bigrams) with the edge weight value
(Paranyushkin, 2020).
B. Topic modeling to ‘tf-idf algorithm
Text mining algorithms can model models of words, Those words are central for meaning
terms, phrases, complete sentences, and complexities. The circulation and can be seen as the meaning junctions
modern text mining approach to textual research is within the discourse — they are shown bigger on the
becoming an innovative style of research in this area. Text graph (Paranyushkin, 2020).
mining is used to analyze text clusters, text macro
structures, sentiment analysis, and conversion of spoken b) Research case:
text into numeric values. It is a “mathematical method and Speech by Ts. Elbegdorj president of Mongolia at
programming language” interface model for any the general debate of the 71th session of the united
presentation, speech, text, or source text. It is an important nations general assembly on “The sustainable
way to process and analyze information across sectors. development goals: a universal push to transform
‘Text mining = Methods for extracting useful information our world” (Elbegdorj, 2006)
from large and often unstructured collections of texts. It is
also closely related to “information retrieval.” In this
context, keywords that carry information about the contents
of a document are called terms. A list of all the terms in a
document is called an index. For each term, a list of all the
documents that contain that particular term is called an
inverted index, a matrix normally represents the frequency
of occurrence of term in document.
(Saito, 2013)

IJISRT22MAY581 www.ijisrt.com 74
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
node name degree frequency betweenness topic diversivity
sum total 1752 688 2.770113 n/a 3787.6
sum / 150 nodes 11.68 4.59 0.018467 n/a 25.25
mongolia 57 25 0.357274 2 142.9
people 42 16 0.174323 2 109
development 38 17 0.170083 3 100
human right 34 19 0.147893 7 77.8
government 22 9 0.1397 3 155.2
world 32 14 0.12048 2 86.1
sustainable 17 8 0.092962 4 116.2
international 27 9 0.07734 6 85.9
law 24 11 0.075753 4 68.9
peace 22 10 0.067855 6 67.9
sustainable 22 9 0.060206 3 66.9
Table 1: Discourse semantic network table

Data table of Semantic macrostructure of “The sustainable development goals: a universal push to transform our world”
(Elbegdorj, 2006)

Frequency: values with semantic significance in the discourse structure: 1752 /150 nodes,
The most Significant values: between topics
“MONGOLIA” 57/25/0.357/2
“PEOPLE’ 42/16/0.174/2
“DEVELOPMENT”38/17/0.17

The network's structure diversity is higher which is its mind-viral immunity — that is, such a network will be more resilient
and adaptive than a less diverse one. In case of a discourse network, high mind-viral immunity means that the text proposes
multiple points of view and propagates its influence using both highly influential concepts and smaller, secondary topics.

Modularity: 0.53 (medium)

Influence Distribution 80%
(the measure of influence distribution (the entropy of the top nodes' distribution among the top
clusters),
as well as the the percentage of nodes in the top community)

Fig.1: Discourse semantic network and topics

The topics are the nodes (words) that tend to co-occur together in the same context (next to each other).
Detection algorithm: a combination of clustering and graph community detection algorithm to identify the groups of nodes are
more densely connected together than with the rest of the network.

IJISRT22MAY581 www.ijisrt.com 75
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

They are aligned closer to each other on the graph are given a distinct color.

The top prominent nodes that have significantly higher influence than the rest.

The most significant nodes are either the ones with the highest betweenness centrality (current setting) — appearing most
often on the shortest path between any two randomly chosen nodes are ‘MONGOLIA’, ‘PEOPLE’, ‘DEVELOPMENT’

The higher is the diversity, the more distinct communities (topics) there are in this network, the more likely it will
be pluralist.

Table II. Shows the percentage of Topics of Fig.1

Fig. 2: Influence betweenness topics

The fig.2 Shows how influence propagates through the network.

X-axis: lemma to lemma step (narrative chronology).
Y-axis: change of influence.

The more even and rhythmical this propagation is, the stronger is the central idea or agenda Propagation dynamics: irregular
variability | alpha exponent: 0.60 | high (based on Detrended Fluctuation Analysis of influence)

IJISRT22MAY581 www.ijisrt.com 76
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 2: Influence fluctuation scale between topics

Propagation dynamics: Irregular variability | alpha exponent: 0.60 | high

(based on Detrended Fluctuation Analysis of influence)

The more variability can be seen in the propagation profile, the less is the reliance on the main concepts (agenda), the stronger is
the role of secondary topical clusters in the narrative.

We plot the narrative as a time series of influence (using the words' betweenness score). We then apply detrended fluctuation
analysis to identify fractality of this time series. Plotting the log2 scales (x) to the log2 of accumulated fluctuations (y).

ks: 1.42, d: 0.37 > cr: 0.35 | power law not identified (based on kolmogorov-smirnov test)

If a power-law is identified, the nodes have preferential attachment (e.g. 20% of nodes tend to get 80% of connections), and
the network may be scale-free, which may indicate that it's more resilient and adaptive. Absence of power law may indicate a
more equalized distribution of influence.

Fig. 3: Influence fluctuation Alpha fit between topics

Using the alpha exponent of the fit (which is closely related to Hurst exponent)),
 relation: uniform (pulsating | alpha <= 0.65),
 variable (stationary, has long-term correlations | 0.65 < alpha <= 0.85), fractal (adaptive | 0.85 < alpha < 1.15), and complex
(non-stationary | alpha >= 1.15).
 "fractal" (near-critical state): maximal diversity, adaptivity, and plurality, the narrative
 "uniform" (For fiction, essays, and some forms of poetry)
 "variable + stationary" (Informative)
 "complex" (state is an indicator or shifting its state)

IJISRT22MAY581 www.ijisrt.com 77
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 4: Fraction of nodes

ks: 1.42, d: 0.37 > cr: 0.35 | power law not identified (based on kolmogorov-smirnov test)

If a power-law is identified, the nodes have preferential attachment (e.g. 20% of nodes tend to get 80% of connections), and
the network may be scale-free, which may indicate that it's more resilient and adaptive. Absence of power law may indicate a
more equalized distribution of influence.

In combination with the measure of influence distribution (the entropy of the top nodes' distribution among the top clusters),
as well as the the percentage of nodes in the top community is below:

Fig. 5: Scaled nodes network

ks: 1.42, d: 0.37 > cr: 0.35 | power law not identified (based on kolmogorov-smirnov test)
(network has scale-free / small-world (long-tail power law distribution) or random (normal, bell-shaped distribution) network
properties.

IJISRT22MAY581 www.ijisrt.com 78
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 5: Evolution of topics in discourse

The chart shows how the main topics and the most influential keywords evolved over time.
X-axis: time period (split into 10% blocks).
Y-axis: cumulative frequency of occurrence.

Fig. 6: K-NN scaling, The nearest semantic neighbors of predicted space

Integrating text: semantic and grammatical rules,
VI. METHODS FOR NETWORK ANALYSIS tools for connecting parts of the text, and their use. Logical
relation of meaning: coherence, the logical sequence in
The Method is an open-source software written in structure, optimal wording style, and the correct choice of
JavaScript using Sigma.Js, Cytoscape and Graphology vocabulary will be considered psychological factors in
libraries in the front-end and java-based Neo4J graph discourse.
database for the backend, and SPSS: K-NN.
A mental model: fact-based knowledge arises from a
VII. SUMMARY basic concept, which is a fundamental issue of language,
behavior, and cognition, and is a complex set of thoughts
Regarding the text structure theory, a theory that and mindsets, (Б.Дагиймаа). Mental models, mental maps
explains the logical connection and coherence of meanings and mind maps are often used to represent complex ideas
in a text-based on the principle that each part of the text and and have a positive impact on teaching and learning, can be
speech has its special function, there are no words or extracted from texts to produce cognitive maps, which can
meanings, a hierarchical meaning structure, and a unified then be used for sentiment analysis. (Paranyushkin, 2020).
introduction and conclusion.

IJISRT22MAY581 www.ijisrt.com 79
Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Topical modeling: Topical modeling is a key method used ACKNOWLEDGMENTS
in text analysis to determine keyword frequency.
This research work was supported by text
VIII. RESULT OF NETWORK ANALYSIS visualization and topic modeling algorithms on nodus
source and SPSS platform. We also want to thank for
The network's structure diversity is higher, which is providing access to their resources for research purposes.
its mind-viral immunity As well, thanks to speech by Ts. Elbegdorj president of
Mongolia at the general debate of the 71th session of the
Modularity: 0.53 united nations general assembly on “The sustainable
Influence distribution: 80% development goals: a universal push to transform our
The topics are the nodes (words) that tend to co-occur world” (Elbegdorj, 2006)
together in the same context (next to each other). REFERENCES
Detection algorithm: a combination of clustering and graph
community detection algorithm to identify the groups of [1.] B.Dagiimaa, & O.Munkhchimeg. (2022). Text Mining
nodes are more densely connected together than with the Applications in Discourse Analysis and Modern
rest of the network. Trends (Bid Data Methods on Political Speeches)
[2.] Bloomfield. (1998). The MIT Encyclopedia of
The most significant nodes are either the ones with Cognitive Science. (F. C. Robert A. Wilson, Хян.)
the highest betweenness centrality (current setting) — MIT PressEditors.
appearing most often on the shortest path between any two [3.] F.Ungerer, & H.J. Schmid. (2012). An introduction to
randomly chosen nodes are ‘MONGOLIA’, ‘PEOPLE’, Cognitive Linguistics. (E. P. M.A.K. Halliday., Хян.)
‘DEVELOPMENT’ Foreign Language Teaching and Research Press.
The higher is the diversity, the more distinct [4.] F.Ungerer, H. S. (2012). An introduction to Cognitive
communities (topics) there are in this network, the more Linguistics. (E. P. M.A.K. Halliday.) . Foreign
likely it will be pluralist. Language Teaching and Research Press.
X-axis: lemma to lemma step (narrative chronology). [5.] Fairclough. (2017). Critical Approach In Social
Y-axis: change of influence. Research. The Online Journal of Communication and
Media.
Propagation dynamics: irregular variability | alpha https://www.tojned.net/journals/tojcam/articles/v03i04
exponent: 0.60 | high (based on Detrended Fluctuation /v03i04-02.pdf-ээс Гаргасан
Analysis of influence) plotting the log2 scales (x) to the [6.] Van Dijk in Wodak & Chilton, 2005, p. 71-72.
log2 of accumulated fluctuations (y). Critical Discourse Analysis.
[7.] Б.Дагиймаа. (2020). Дискурсын онол арга зүйн
ks: 1.42, d: 0.37 > cr: 0.35 | power law not асуудлууд: ярилцлагын дискурсын жишээгээр (Б.
identified (based on kolmogorov-smirnov test) 4/2020). Улаанбаатар хот: Европ судлал, эрдэм
relation: uniform (pulsating | alpha <= 0.65), шинжилгээний сэтгүүл.
[8.] Staggers, Nancy; Norcio, A.F. (1993). "Mental
variable (stationary, has long-term correlations | 0.65 models: concepts for human-computer interaction
< alpha <= 0.85), fractal (adaptive | 0.85 < alpha < 1.15), research". International Journal of Man-Machine
and complex (non-stationary | alpha >= 1.15). Studies.
"fractal" (near-critical state): maximal diversity, adaptivity, [9.] M. Young, The Technical Writer’s Handbook. Mill
and plurality, the narrative. Valley, CA: University Science, 1989.
[10.] Paranyushkin, D. (2020). InfraNodus: Generating
This discourse's structure is balanced representation of Insight Using Text Network . Paris, France.
several perspectives and its immunity level is high. [11.] Elbegdorj, T. (2006). “THE SUSTAINABLE
It can be made more focused if develop the ideas DEVELOPMENT GOALS:. Ulaanbaatar, Mongolia:
around the main keywords — MONGOLIA, PEOPLE, https://www.un.int/mongolia/statements_speeches/.
DEVELOPMENT, RIGHT — or by identifying the
structural gap between the less represented topics
X-axis: time period (split into 10% blocks).
Y-axis: cumulative frequency of occurrence.
The nearest semantic neighbors of predicted space,
identified by the factor and variables with K-NN SPSS

IJISRT22MAY581 www.ijisrt.com 80