Advances in Intelligent Systems Research, volume 174

Proceedings of the 8th Scientific Conference on Information Technologies for

Intelligent Decision Making Support (ITIDS 2020)

Analyzing the Complex Nature of Emotions From

Written Text Using Parrott Theory of Emotions

Mohsin M. Abbasi Anatoly Beltiukov*

Theoretical Foundation of Computer Sciences Theoretical Foundation of Computer Sciences
Udmurt State University Udmurt State University
Izhevsk, Russian Federation Izhevsk, Russian Federation
mohsinmanshad@gmail.com belt.udsu@mail.ru

Abstract— Emotions are the main component of a person’s other, while others are partially dependent, with few of them
character and personality. They are used to communicate simply independent [1].
messages and information. Emotions are represented in various
ways, such as by physical actions, verbal tone, written text, etc. In our previous works, we observed a relationship between
We are interested in identifying the emotions presented in the different emotions depending on the index of their occurrence
written text and then to analyze them. Emotions are complex in and the distance between them in the text [2-5]. In this paper,
nature. An emotion exhibits its own nature and characteristics. we analyze the nature of emotions and the relationship
It may represent another emotion in a written text. The complex between them based on their characteristics in the written text
property of emotions is similar to the characteristics of complex document.
numbers, which are represented by their real and imaginary
parts. Various algorithms are used to identify and analyze The proposed methodology helps to analyze the complex
emotions in the text. These techniques usually considered each characteristics of emotions and to understand the real and
emotion as a separate element with its own characteristics. The imaginary parts of them. For example, let's say “Andrey loves
relationship between different emotions and their influence on Lena”. Here the only emotion explicitly presented in the
the overall polarity of the text has not received much attention sentence is love. However, it also means that he likes to spend
from researchers over a period. In this work, the main idea is to time with her. He feels happy and shows affection towards
analyze the complex characteristics of emotions by identifying her. Here the emotions “happy” and “affection” are
the relationships and dependencies that exist between them. To imaginary emotions of the true emotions “love”. These
support this idea, theories of emotions such as Plutchik's wheel emotions are not presented in the text, but they do exist in the
of emotions and Parrott's theory of emotions are utilized. meaning. They are dependent semantically and logically on
According to these theories, emotions are classified into their primary emotion love as presented by Parrott in his work
different groups and levels depending on their intensities and during 2001 [6].
dependencies. The results of the research are described in detail
in the later sections of the article. Similarly, Eugene is afraid of the exam. Here the real
emotion expressed in the text is “afraid”. However, there are
Keywords—Parrott theory, emotion, class, summarization, imaginary emotions associated with it, such as panic, stress,
internet, blogs, communication horror, tension, etc. Here we can see that an emotion
expresses the existence of several nonexistent emotions in the
I. INTRODUCTION text, or an emotion can represent many other emotions.
Text analysis over the past decades considered a main area
II. HISTORY OF EMOTION ANALYSIS
of research. It is used as a mechanism for identifying and
transmitting information over the Internet. The popularity of The first system that was developed to detect emotions in
text analysis is due to the growth of the Internet. The Internet the text was the Inquirer system in the 1960s at the Harvard
allows the persons to express their ideas, emotions, beliefs, Laboratory [7]. The main goal of the system was to identify
opinions etc., more efficiently. Today, one of the important positive and negative emotions in the text. After that, there
sources of accessible text is the Internet. There is a huge was a slight delay in the research until the beginning of the
amount of data on the Internet, available in the form of social 90s, when the term “Subjectivity” was firstly introduced for
networks, interactive groups, blogs, etc. The text over the research in information retrieval systems [8].
internet is available in different languages and is growing at a
At the beginning of the 21st century, Peter Turney
very fast pace.
proposed the idea of "Thumbs up" and "Thumbs down" to
People use the internet to express their opinions and present the positive and negative reviews in the text before
emotions. A text available on the Internet is analyzed to obtain their classification [9]. In 2002, Pang manually created a
useful information from it, such as a forecast of future events, vocabulary that expressed feelings in the text [10]. In 2009,
public polls, weather forecasts, people's opinions on an interesting study was published on several domains to
governmental and organizational policies, etc. In general, illustrate the importance of polarity indicators from the well-
emotions expressed in a text, paragraph, or sentence are known SentiWordNet database [11]. It was noted that the
related to each other. Some are highly dependent on each preprocessing process of the text before analyzing emotions
from it improves the results of the text analysis.

Copyright © 2020 The Authors. Published by Atlantis Press B.V.

This is an open access article distributed under the CC BY-NC 4.0 license -http://creativecommons.org/licenses/by-nc/4.0/. 99
 Advances in Intelligent Systems Research, volume 174

Among the various machine learning and data level to the emotions of another level. Primary emotions are
classification algorithms, clustering appears to be more the most intense emotions, while tertiary emotions are the
effective for the purpose of classification of text into different least intense emotions. The relationship between emotions
clusters. The process of clustering requires a special distance observed below in table 1.
function for creating the clusters of text and to automatic TABLE 1 Parrot Theory of Emotions 2001
determine associations, similarities between different
members of them [12]. Clustering has various types, such as
K-means, Hierarchical, Fuzzy C-means, etc. For
classification purposes, C-means and K-means are apparently
more efficient and effective than others. These methods use
the Euclidean distance function to divide emotions into
different groups. To determine the semantic orientation of
text, adjectives are seen as an important measure.
Emotional modalities and their logical formulas were
used to analyze the text and its orientation. Logical models
classify text emotions in different groups based on their
hereditary nature [13-15]. Analysis of emotions and moods in
text written in the Russian language began in the last decade.
Initially, the already used templates and techniques in other
languages were applied directly to the text written in Russian
language. However, the difference in the writing style and
syntactic features of the Russian language did not allowed
existing models to analyze emotions from the text correctly
and accurately [16].
In 2009, the analysis was conducted on a blog of the
Russian automobile community for the analysis of emotions
to determine the mood of people regarding a particular brand
of car in the Russian market [17]. An experiment was
conducted on a text in Romanian and Russian to find out if
the spelling of their words corresponds to an equivalent
phonetic word or not. The result of the experiment was used
to improve the capability of the search for words in the
written text [18].
Similarly, an experiment was conducted on English and
Russian language reviews about the same books to observe
the relationship between emotions in both languages [19].
Loukachevich and Chetviorkin proposed a mechanism for
identifying Russian sentiments regarding reviews of goods
and services [20]. Then Steinberger et al. built a dictionary of
emotions in different languages [21]. In 2014, Poria et al. used
conceptual rules in tree-structure to identify sentiments in text
document [22]. According to Parrott’s theory, the emotions are classified
into three groups or levels named as primary, secondary and
Rule-based methods seemed to be a very attractive tertiary levels. Primary emotions are independent emotions.
technique for creating grammar, and then for performing The secondary emotions are dependent upon primary
parsing and semantic analysis of text written in Russian emotions whereas the tertiary emotions are dependent
language. However, the effectiveness of machine learning directly upon secondary emotions and indirectly upon
methods for categorizing and classifying emotions from text primary emotions.
outclassed the rule-based mechanisms. In recent years, the
syntax analysis has attracted more attention than the It means while allocating the sentiment score to each
semantics analysis of text for analyzing emotions from it. emotion, it is required to observe its relationship to other
emotions and the dependencies that exist between them. The
III. METHODOLOGY AND DISCUSSION complex property of emotions is that emotion may exhibit the
The purpose of this study is to identify the complex properties of some other emotions or it may represent other
characteristics of emotions in the written text, such as the emotions in text. Based on this hypothesis, the polarity scores
relationship between them, their dependencies, etc. For this are allocated to emotions using the equations below.
purpose, we use the Parrott’s Theory of Emotion to classify 𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒(𝐴) = 𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒(𝐴) + ∑𝑁
𝑖=1 𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒 (𝐵(𝑖)) (1)
emotions into different groups.
According to him, emotions can be classified into three 𝑖=1 𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒 (𝐶 (𝑖))
𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒(𝐵) = 𝑝𝑜𝑙𝑠𝑐𝑜𝑟𝑒(𝐵) + ∑𝑁
different levels or groups depending on their nature and polscore(C) = polscore(C)  
intensities, such as primary, secondary, and tertiary emotions.
A unique relationship exists between the emotions of one

100
 Advances in Intelligent Systems Research, volume 174

From the above equations, it can be observed that the negative emotions in the text. A part of the text from online
polarity value of an emotion belonging to the main level A is blog is presented below in Fig. 2.
the sum of its own polarity value with the polarity values of
all its dependent emotions from level B and C. Similarly,
polarity value of an emotion belonging to level B is the sum
of its own polarity value with the polarities of all its
dependent emotions in level C. whereas the emotions in level
C have just their own polarity values.
In other words, the polarity score of real emotions are
their own polarity, added with the polarity score of all their
imaginary emotions. The results of the above equations be
visualized using the figure below.

Fig. 1 Logical relationship between different emotions according

to Parrott’s theory of Emotions

In Fig. 1 above, state C represents tertiary emotions,

presented in table 1. Emotions at this level are usually short-
term and mild emotions. If they exist for a longer period, they
Fig. 2 Blog of text from interne about online shopping
move to a higher and extreme level of emotions in state B.
Emotions in state B are secondary emotions. They have their
From the above text, we identified the words that express
own characteristics plus the characteristics of tertiary
emotions in text. Those words (emotions) were then analyzed
emotions that follow them. When secondary emotions exist
using Parrott's theory of emotions for their classification and
at this level for a longer period, they move to the highest and
updating of polarity scores. We applied our proposed
most intense level of emotions called primary emotions.
methodology to calculate the intensities of emotions based on
Primary emotions are intense and long-lasting emotions.
different levels of intensities of Parrott’s theory. For making
These are mainly irreversible emotions, and in the case to
the calculations simple, a polarity score of 1 is assigned to the
reverse them, they require a longer period. Primary emotions
emotions of tertiary level, 2 to secondary emotions, and 3 to
contain their own properties, as well as the properties of their
the emotions of primary level in the text. The experiment and
secondary and tertiary emotions.
its results are detailed below in table 2.
The above figure details the relationship between
emotions based on the categories of emotions proposed by TABLE 2 Emotions and their Intensities in text using Parrot Theory
Parrott in his theory. We can observe that real emotions, as in
the case of primary emotions, have the properties of their Emotions and their Intensities using Parrot Theory of Emotions
imaginary emotions, such as secondary (level B) and tertiary
(level C) emotions.
Occurrences

intensities
Including
Emotion

Intensity

Intensity
Polarity

To determine the emotions and their dependencies, each

Overall
Factor
score

other

sentence from the text was analyzed separately in the

experiment. When tertiary emotions explicitly exist in the
text, they contain only their own estimated polarity. While
secondary emotions in the text contains all the characteristics Happy 1 1 1 - 1
and polarity values of their tertiary emotions (imaginary Satisfied 2 1 2 - 2
emotions), as well as their own (real emotions) polarities. Secure 1 1 1 - 1
Similarly, primary emotions contain their own polarity values 2 1 2 - 2
Safe
along with polarity values of their secondary and tertiary
Trust 1 1 1 - 1
emotions (imaginary emotions).
The above methodology helps to understand the logical Easiest 3 1 3 - 3
relationships that exist between different emotions in a text. Surprise 2 3 6 5 30
This relationship between emotions can be used to update the Fear 1 3 3 9 27
polarity of each emotion in the text and to determine the Risk 3 1 3 - 3
overall strength of the emotion in the text.
Fake 2 1 2 - 2
IV. EXPERIMENT & RESULTS Scams 1 1 1 - 1
For the experiment, a text blog from the Internet was Significant 1 1 1 - 1
downloaded. The text contains an opinion on online Affection 1 2 2 10 20
shopping. Words with green color in figure 2 represents Love 1 2 2 16 32
positive emotions, while words with red color represents

101
 Advances in Intelligent Systems Research, volume 174

It was observed that the emotions have different number Technology Journal. Academy of Telecommunications and
Informatics, Samara, 2019. vol. 2. issue 17, pp. 246-254. (Аббаси
of occurrences with different intensities in the blog of text. M.M., Бельтюков A.П., Хусейн Лалл., Aббаси A.К. “Анализ
Using the Parrott’s theory of emotion, the intensities of the эмоций из текстов для управления обществом”. Journal of
emotions in the blog of text were calculated. According to the Инфокоммуникационные технологии журнал. Академия
proposed methodology, independent emotions should have a телекоммуникаций и информатики, Самара, 2019. vol. 2. issue 17,
pp. 246-254).
higher polarity value than their dependent emotions. This
[5] Abbasi M.M., Beltiukov A.P. “Identifying the strength of emotions in
concept is the main idea of the proposed methodology. The relation with the topic of text using Word space”, In the Proceedings of
emotions from text contain their own value of polarity along the 21th International workshop on computer science and information
with the polarities of their dependent or imaginary emotions. technologies, Austria, Vienna, 2019. Journal of Atlantis Highlights in
The analyzed emotions from text with their intensities are Computer Sciences. issue. 3. pp. 1-5.
plotted using the graph below. [6] Parrott, W. Emotions in Social Psychology, Key Readings in Social
Psychology. Philadelphia: Psychology Press. ISBN 978-0863776830.
2001.
[7] Stone, P.J, Dunphy, D.C, Smith, M.S. “The General Inquirer: A
Computer Approach to Content Analysis”. MIT Press - Cambridge,
35 1966.
30 [8] Wiebe, Janyce M. “Identifying Subjectivity characters in Narrative”,
Intensity of Emotion

In Proceeding of the 13th International Conference on Computational

25 Linguistics, Helsinki Morristown NJ: Association of Computational
20 Linguistic, 1990, pp. 401-408.
[9] Peter D.T. “Thumbs Up or Thumbs Down? Semantic Orientation
15 Applied to Unsupervised Classification of Reviews”, In the
10 Proceedings of the 40th Annual Meeting of the Association for
Computational Linguistics, Philadelphia, 2002, pp. 417-424.
5
[10] Pang B, Lee L. “Thumbs up? Sentiment Classification using Machine
0 Learning Techniques”, In the Proceedings of the Conference on
Fast
Safe

Surprise
Secure

Risk
Fake
Satisfied

Scams
Trust

Significant
Fear
Happy

Easiest

Affection
Love

Empirical Methods in Natural Language, Philadelphia, 2002, pp.79-86.

[11] Denecke K. “Are SentiWordNet scores suited for multi-domain
sentiment classification?”, In the Proceedings of the fourth
Emotion International Conference on Digital Information Management,
Intensity in Text University of. Michigan, USA, 1-4 November 2009.
[12] D. Boley, M. Gini, R. Gross. “Partition based clustering for web
Fig. 3 Emotions and their intensities in Text document categorization”. Elsevier Journal for Decision Support
Systems, vol. 27, issue 3, December 1999, pp. 329-341.
The graph above shows the primary emotions in the text [13] Lellmann B., Pattinson D. “Sequent Systems for Lewis’ Conditional
contain their own characteristics, polarities along with the Logics”. In: del Cerro L.F., Herzig A., Mengin J. (eds) Logics in
Artificial Intelligence. JELIA 2012. Lecture Notes in Computer
polarities and the characteristics of their imaginary emotions. Science, Springer, Berlin, vol 7519.
[14] W.A. Carnielli and C. Pizzi. “Modalities and Multimodalities. Logic,
V. CONCLUSION Epistemology, and the Unity of Science” Springer, Berlin, vol. 12,
We observed a logical relationship between the emotions 2009, pp. 312.
that exist in the text. Emotions that exist in the text represent [15] Abbasi M.A, Beltiukov A.P. “Logical analysis of emotions in text from
several other nonexistent emotions. These nonexistent natural language”. Vestnik Udmurtskogo Universiteta.Matematika.
Mekhanika. Komp'yuternye Nauki, vol. 29, issue 1, 2019, pp. 106-116.
emotions are imaginary emotions of the real emotion. While
identifying the intensity of an emotion, it is important to [16] Pivovarova L, Du M, Yangarber R. “Adapting the PULS event
extraction framework to analyze Russian text”, In the Proceedings of
observe its logical characteristics. Emotion and its logical the 4th Biennial International Workshop on Balto-Slavic Natural
characteristics can affect the overall emotional polarity of Language Processing, Sofia, Bulgaria, 8-9 August 2013, pp. 100–109.
text. It was observed that by considering the complex nature [17] Ermakov A. “Knowledge extraction from text and its processing:
of emotion helps to analyze the emotional subjectivity of the Current state and prospects”, In the Proceedings of the Computational
text. Using this methodology, the emotions can be effectively Linguistics and Intellectual Technologies, 2009. pp. 50–55.
summarized from the text. [18] Sokolova M, Bobicev V. “Classification of emotion words in Russian
and Romanian languages”. International Conference RANLP,
REFERENCES Borovets, Bulgaria 2009, pp. 416–420.
[19] Zagibalov, T , Belyatskaya et al. “Comparable English-Russian Book
[1] Plutchik, R. The Nature of Emotions. Archived from the original on Review Corpora for Sentiment Analysis”, Edition 2010.
July 16, 2001.
[20] Chetviorkin I, Braslavskiy P, Loukachevich N. “Sentiment Analysis
[2] Abbasi M.M, Beltiukov A.P. “Summarizing Emotions from Text
Track at ROMIP 2011” , Computational Linguistics and Intellectual
Using Plutchik’s Wheel of Emotions”, Advances in Intelligent systems
Technologies. In the Proceedings. of the International Conference
research. In Proceedings of the 7th Scientific Conference on (Dialog 2012 ), Bekasovo, pp. 1–14.
Information Technologies for Intelligent Decision Making Support
(ITIDS 2019), Atlantis press, vol. 166, pp. 291-294. [21] Steinberger J, Lenkova P, Kabadjov M., et al. “Multilingual Entity-
Centered Sentiment Analysis Evaluated by Parallel Corpora”, In the
[3] Abbasi M.M. “Analysis of the influence of negation on the text using
Proceedings of Recent Advances in Natural Language Processing,
clustering and Plutchik wheel of emotions.” Scientific and Technical
Bulgaria, 12-14 September 2011, pp. 770–775.
Journal of the Volga Region, Kazan, 2019. vol.66, issue 12, pp. 23-28.
(Аббаси M.M. “Анализ влияния отрицаний на текст с [22] Poria S, Ofek N, Gelbukh A, Hussain A, Rokach L. “Dependency Tree-
использованием кластеризации и колеса эмоций плутчика”. Based Rules for Concept-Level Aspect-Based Sentiment Analysis”, In:
Journal Научно -технический вестник Поволжья, Казань, 2019. Presutti V. et al. (eds) Semantic Web Evaluation Challenge.
vol.66, issue 12, pp. 23-28). SemWebEval 2014. Communications in Computer and Information
Science, vol 475. Springer, Cham.
[4] Abbasi M.M., Beltyukov A.P., Hussain Lal., Abbasi A.Q. “Analysis of
emotions from texts for managing society. ” Infocommunication

102