A method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions / word forms in a text containing multiple words and its intensity, as well as a method and system for such detection

Abstract

The invention relates to the field of analytical review of texts, in particular a method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions in a text including multiple words and its intensity, as well as a method and system for such detection, and can be used, for example, to assess the emotional coloring of news covered in the media and intensity of their emotional impact on the audience. The claimed invention ensures achieving a technical result consisting in the ability to thoroughly assess the emotional coloring of the text and its intensity.

Description

[0001] A METHOD FOR TRAINING A MACHINE LEARNING MODEL TO AUTOMATICALLY DETECT THE EMOTIONAL COLORING PECULIAR TO EXPRESSIONS / WORD FORMS IN A TEXT CONTAINING MULTIPLE WORDS AND ITS INTENSITY, AS WELL AS A METHOD AND SYSTEM FOR SUCH DETECTION

[0002] The invention relates to the field of analytical review of texts, in particular a method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions in a text including multiple words and its intensity, as well as a method and system for such detection, and can be used, for example, to assess the emotional coloring of news covered in the media and intensity of their emotional impact on the audience.

[0003] Currently, there are several methods for analyzing the emotional coloring of a text.

[0004] A method aimed at detecting aggression and "cyberbullying" in social networks has been researched by examining keywords and word combinations by such researchers as Aditya Malte and Pratik Ratadiya. Using the BERT bidirectional transformer architecture, they researched Facebook texts in two languages: English and Hindi (Acheampong, FA, Wenyu, C, Nunoo-Mensah, H. (2020). Text-Based Emotion Detection: Advances, Challenges, and Opportunities. Engineering Reports; 2:el2189. [Online], Available: https: / / doi.org / 10.1002 / eng2.12189 (accessed 23 September 2021)) [1],

[0005] Detection of emotions by using existing emojis in the text has been proposed by researchers Chen and LeCompte. They used the Keyword Detection Method, the Support Vector Machine and the Multinomial Naive Bayes Method, which yielded a classification of emotions into sadness, anger, fear, pleasure, surprise, gratitude, and love. This approach improved detection capabilities.

[0006] Automatic data detection in a multilingual text has been reported in the study [1] by Jian et al. The proposed structure uses the Natural Language Processing (NLP) method to classify selected words based on Ekman's concept (of emotion models) by using first the SVM (Support Vector Machine) method and then NB (Naive Bayes).

[0007] In [1], the development of an automatic emotion classifier for processing the Twitter network is described, where tweets are analyzed by using three text corpora of emotions,1where features are first extracted from these databases individually, and synonyms of words denoting emotions are generated by using the WordNet dictionary in the databases. The model was trained by using the SVM classifier, and emotions were grouped under one of the 6 categories of Ekman’s model.

[0008] The Emotex model has been developed to analyze the Twitter social network, in particular to properly assess the audience's sentiments (Hasan, M., Elke, A., Rundensteiner, E.Agu. (2014). EMOTEX: Detecting Emotions in Twitter Messages. [Online] . Available: https: / / web.cs.wpi.edu / ~emmanuel / publications / PDFs / C30.pdf (accessed 23 September 2021)) [2], which allows to detect emotions in a text by using the teacher-assisted learning method and emotion dictionaries. The approach includes 2 methods for performing classification tasks with and without distribution of lexical items on the Internet. The Emotex model is based on texts from the Twitter social network with labelled emotions and SVM, NB classifiers and the Decision Tree. The data are pre-processed and divided into structures according to the vectors, according to which training sets (datasets) for a classification model are created as a result. When working with Twitter, real-time classification of tweets on the Internet uses datasets and processing processes available offline. The model has a simple structure and describes a wide range of emotional states denoted with 28 main words. It is compatible with the Circumplex model, the emotions found are distributed on a two- dimensional scale, where they are assessed according to the characteristic "satisfied" - "dissatisfied" and the degree of activity: "satisfied active" - "satisfied inactive", and "dissatisfied active" - "dissatisfied inactive".

[0009] The review of existing approaches shows that there is a lack of an appropriate analytical apparatus that would take into account various cases of the use of word forms that affect the formation of the emotional coloring of the text, a narrow selection of emotions and emotional states that can be detected with the available tools, and a lack of a mechanism for analyzing the interaction of word forms that affect the formation of emotional coloring.

[0010] In addition, all of the listed approaches lack a thorough assessment of the emotional characteristics of tlie text. This particularly applies to the assessment ofthe properties of the texts of the news covered in the media, which shape public sentiments with subsequent consequences. The existing tools are unable to detect a wide range of emotions, which are studied in modem research in the field of psychology and are considered an important characteristic of human being. Also, the listed models do not allow to define such emotions as humor and irony as a derivative of its manifestation.

[0011] Thus, the purpose of the claimed invention is to develop a method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions / word forms in a text and its intensity, a method for automatically detecting the emotional coloring peculiar to expressions / word forms in a text and its intensity, as well as a system implementing said method, which would ensure achieving a technical result consisting in the ability to thoroughly assess the emotional coloring of the text and its intensity, in particular to measure the level of empathy of the text.

[0012] The purpose is solved by developing a method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includes creating a database of emotions, each of which is assigned a weight coefficient of emotional coloring intensity; creating a database of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; creating a database of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; creating a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; creating a training data set by using the created databases; training a model by using the created training data set to obtain a trained machine learning model.

[0013] In addition, the purpose is solved by developing a method for automatically detecting the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includes receiving a text and dividing it into sentences; analyzing the text by using a machine learning model with the labelling of its emotionally colored words / word combinations / word forms with one of the emotions and its weight coefficient; wherein the machine learning model is trained by using a training data set created by using databases: of emotions, each of which is assigned a weight coefficient of emotional coloring intensity; of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; analyzing the text obtained after performing the previous step by using the rules of syntactic subordination with the recognition of language constructs / word forms within the text and the interpretation of selected emotionally colored words / word combinations / word forms; visualizing the analysis results. Also, the purpose is solved by developing a system for automatically detecting the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includes at least one processor, at least one machine-readable medium communicatively coupled to the at least one processor, and program instructions for detecting the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, stored on the at least one machine-readable medium and executed by the at least one processor, which include program instructions for receiving a text and dividing it into sentences; a machine learning model trained by using a training data set created by using databases: of emotions, each of which is assigned a weight coefficient of emotional coloring intensity; of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; program instructions for analyzing the text by using the machine learning model with the labelling of its emotionally colored words / word combinations / word forms with one of the emotions and its weight coefficient; program instructions for analyzing the text obtained after performing the previous step by using the rules of syntactic subordination with the recognition of language constructs / word forms within the text and the interpretation of selected emotionally colored words / word combinations / word forms; program instructions for visualizing the analysis results. The creation of the above-listed databases for the purpose of subsequent training of a machine learning model, including a database in which a wide range of emotions can be entered, in particular those that are studied in modem research in the field of psychology and are considered an important characteristic of a human being; a database of words / word combinations / word forms, each of which is assigned a corresponding emotion and its weight coefficient of intensity; as well as a database of logical and semantic rules (LSR), which determine the interaction of words / word combinations / word forms within a sentence to obtain an appropriate result, allows to correctly and more accurately determine the emotional coloring of a text.

[0014] The method and system as claimed essentially "run" a text through a machine learning model, including by using LSR, which, as mentioned above, determine the interaction of words / word combinations / word forms within a sentence to obtain appropriate results, and at the next step, analyzing said text by using the rules of syntactic subordination (RSS), which take into account the syntactic structure of the sentence and the peculiarities of interaction and ensuring cohesion between its members.

[0015] Thus, the ability to determine a wide range of emotions for the classification of words / word combinations / word forms, the peculiarities of interaction of words / word combinations / word forms in a sentence and the peculiarities of interaction and ensuring cohesion between its members allow to obtain a technical result consisting in a thorough assessment of the emotional coloring of a text and its intensity.

[0016] Preferably, the analysis results are visualized by labelling the emotionally colored words / word combinations / word forms, and / or by constructing a graph, and / or by constructing a diagram.

[0017] The emotionally colored words / word combinations / word forms within the processed text are labelled with an appropriate color that corresponds to a specific emotion.

[0018] The graph can show the distribution of the emotional character of the text and the dynamics of the emotional coloring.

[0019] In the diagram, the colors can show the ratio of the emotions detected in the text, which are contained in the emotionally colored word forms.

[0020] Mainly at the step of text analysis, by using a machine learning model, a database of emotionally colored words / word combinations / word forms, a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence, and a database of logical and semantic rules of interaction of words / word combinations / word forms in a sentence are additionally used.

[0021] Also, mainly at the step of text analysis, by using the rules of syntactic subordination, a database of emotions, each of which is assigned the weight coefficient k of emotional coloring intensity, and a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence are additionally used.

[0022] Thus, text analysis is most preferably performed both by using a machine learning model trained by using a training data set created by using the above-listed databases, and directly by using said databases.

[0023] In addition, at the steps of the analysis, digital coding is used, an example of which will be given below. Obviously, said coding can be modified without departing from the spirit of the invention.

[0024] Said inventions can be implemented within the framework of a global system working with texts.

[0025] An example of such a global system will be explained in more detail by using the following drawings:

[0026] Fig. 1 is a block diagram of a first application of a global system working with texts and including a system for automatically detecting the emotional coloring as claimed;

[0027] Fig. 2 is a block diagram of a second application of a global system working with texts. The global system in which the claimed system can be used ranks texts (for example, news), depending on their emotional coloring and topic. It consists of two applications that work independently of each other.

[0028] Fig. 1 is a block diagram of the operation of a first application of a global system, which provides for a constant news search, collection (downloading their text), determination of emotional coloring and topic, and saving information.

[0029] It scans specific websites and downloads news for a specified period. News texts are downloaded according to the following scheme: obtaining 1 a list of news links from RSS channels, obtaining 2 news texts from the links, classifying 3 the texts by emotional coloring and classifying 4 the texts by topic, determining 5 the importance of the news based on emotional coloring and topic, saving 6 the result to a database.

[0030] Fig. 2 is a block diagram of the operation of a second application of a global system, which performs news processing. It is based on a client-server architecture (a client 7, a server 8, a database 9), and as a result of its operation, the downloaded and saved news sorted by the indicator of emotional coloring are viewed.

[0031] The second application operates according to the following scheme: a client 7 requests articles from a server 8, after which the server 8 requests information about the articles from a database 9. The database 9 provides the server 8 with information about the articles, and the latter, in turn, provides the client 7 with statistical data and the most important news.

[0032] A detailed description of one of the most preferred embodiments of the claimed invention is given below.

[0033] 1) A database of emotions, each of which is assigned the weight coefficient k of emotional coloring intensity is created as follows.

[0034] First, a list of emotions compiled in advance based on the data of research in the field of psychology is created, with a numerical and letter code, an example of which is given in Table 1 below. The purpose of the code will be explained in more detail below.

[0035] Table 1 - List of emotions

[0036]

[0037] Next, to obtain the weight coefficient (k) of intensity for each emotion from the list, at least one Group 1 of experts is gathered. Each of the experts is given a list with the names of emotions, tables for rating the properties of these emotions with an appropriate list of properties and bipolar scales for assessing the properties according to the semantic differential method. Assessment is carried out on the basis of EPA factors (Evaluation, Potency, Activity, which denote evaluation, potency, activity). By defining numerical indicators according to various factors and scales, the persuasive effect of the text can also be identified. The closer the indicators are to the poles of the scales, the greater the impact on the audience according to the selected factors.

[0038] After the experts have rated according to the specified sample, the final rating of each of the emotions is calculated by calculating the weighted average based on the results of the calculation of the property ratings given by the experts of Group 1 at a particular step for each of them. The obtained values are the values of the weight coefficients k of intensity of the relevant emotions. Below in Table 2, they are entered in the 5th column "Weight coefficient k" and grouped by their modulus from |1| to |8|.

[0039] Table 2 - Database of emotions, each of which is assigned the weight coefficient k of emotional coloring intensity

[0040] The numerical code marked with (*) will be needed for further coding during text processing with a program and by using the already created databases in its operation, which will be described in more detail below.

[0041] 2) A database of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity is created as follows.

[0042] At least one Group 2 of experts who are fluent in the language required to analyze a specific group of news texts in that language is involved. The experts are required to label the texts (mark in the texts of the group the emotionally colored words / word combinations / word forms, the meaning load of which makes an impression, evoking a positive / negative emotional response, that is, which are decisive for the automated assessment of texts in the given language) with a corresponding letter code (a letter code from Tables 1 and 2). This step is performed intuitively, based on the perception by the person who assesses. After such labelling, the labelled texts can be checked to verify the assessment and correct it to avoid subjectivity. After completion of labelling, said words / word combinations / word forms are entered into the base of words / word combinations / word forms with the corresponding weight coefficient k, which corresponds to the emotion assigned by the expert. A fragment of such a database in the form of Table 3 is given below as an example.

[0043] Table 3 - Fragment of a database of emotionally colored words and word combinations

[0044] The number of Groups 1 and 2, as well as the number of people in each group, are determined according to the needs of a particular research.

[0045] Thus, the letter code allows to mark words / word combinations / word forms with the corresponding letter code of the emotion to which the lexical unit corresponds, without wasting the expert's time on searching for the numerical value k, and also indicates the degree of positivity / negativity of the word / word combination / word form as the carrier of this emotion.

[0046] Next, data are determined for the classification of sentences according to their meaning by semantic categories, such as object of emotion, subject of emotion etc., and logical and semantic rules during automated text assessment. Table 4 below provides the numerical code of the semantic category of the sentence and the character of its meaning according to the rules (column 1), the semantic category of the sentence and the indicator of the type of interaction of its constituents (column 2) and the code indicators (+ or -) for the words that are in the first or the second positions after the * mark (columns 3 and 4).

[0047] In the columns of Table 4, these + / - indicate the presence or absence of a feature in the constituents of a word form. The "+" sign means that the positioning of a word after * equals the tabular value of 1-8 of emotional coloring intensity (which can be both positive (+) and negative (-)). The sign in the column of "presence of the characteristic" of Table 4 means that the positioning of a word after * (in the first or second place after *) corresponds to the value of "0" and is not assessed. The symbols of columns 1, 3, 4 are intended for training a model so that it recognizes the semantic categories of column 2 when processing the texts subsequently. These categories are linked to the meaning of emotionally colored words / word combinations / word forms and allow to divide sentences into meaningful fragments, which can be used to trace the meaning throughout the text.

[0048] Table 4 - Form of presenting digital coding of semantic categories and logical and semantic rules

[0049] 3) A database of LSR for interaction of words / word combinations / word forms in a sentence is created based on performing many procedures for analyzing the interaction of a large number of words, word combinations and sentences in texts with emotional coloring. An example of the created database is given below in Table 5. These rules help to obtain a more accurate assessment of both sentence fragments and the whole text at the lexical level of organization.

[0050] Table 5 - Database of logical and semantic rules of interaction of word forms in a sentence

[0051]

[0052] Column 3 of Table 5 describes the content of column 2 of Table 4.

[0053] 4) Databases of procedures for processing words / word combinations / word forms and text and of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence are obtained as follows.

[0054] The rule of denial (71*, Table 5) leads to the formation of Table 6 of pairs of antonyms of emotions to determine the character of a word form, whether positive or negative depending on its context.

[0055] Table 6. - Table of pairs of antonyms of emotions

[0056] In addition, there are1cases when interpretation of the emotional coloring of a phrase / information depends on the party that conveys it. This creates the rules of "false bottom", when the same phrase can be perceived by the audience both positively and negatively, depending on the views of this audience and its attitude towards the source of information.

[0057] The rules of "false bottom" include: the rules of false bottom for the context, which are closely related to the awareness and views of the target audience; the rules of false bottom for the source, which determine the emotional coloring of certain word forms of the news depending on the political position of the publication.

[0058] Therefore, the above rules are the main rules for assessing the character of a text and are based on word forms that represent the lexical level of organization of the text.

[0059] In addition, a correct recognition of syntactic constructs ensured by observing the rules of syntactic subordination is important.

[0060] The syntactic principles (SP) of the following order apply to these rules: syntactic rules of ensuring cohesion; syntactic rules of control; syntactic rules of conjunction.

[0061] The use of the mentioned rules (LSR and SP) ensures the assessment of the character of a text according to certain word forms that represent the lexical level of organization of the text, and their effect is ensured by the use of the syntactic level of organization of the text, which takes into account the syntactic structure of a sentence with SP rules and the peculiarities of interaction of the sentence members with the ensuring of cohesion between word forms at the syntactic level of representation of the text.

[0062] Tables 2-6 are used to train an unsupervised machine learning model, as well as at the steps of text analysis.

[0063] In order to detect the emotional coloring peculiar to the expressions of a certain text and its intensity, this text is analyzed by using a machine learning model with the labelling of its emotionally colored words / word combinations / word forms with one of the emotions and its weight coefficient. Tables 3, 4, 5 and 6 are also used at this step. In particular, in each sentence of the text, all the verbs and nouns are detected, a search for the words of the sentence is performed in the database of emotionally colored words / word combinations / word forms (Table 3), the detected word forms with emotions that are in the same field are filtered, leaving longer sequences, taking in account the emotional coloring intensity of the sentence (by using Table 5), the emotions of the sentence and their statistics are added and saved in the database that contains emotionally colored texts analyzed in this process, the emotional coloring intensity of the text is calculated (by using Table 5).

[0064] Statistics refers to a numerical indicator of how many word forms with certain emotions have been detected in a given sentence and to the quantitative ratio of the emotionally colored word forms to the total number of words in a sentence. This allows to calculate the emotional coloring intensity of the sentence and subsequently of the whole text.

[0065] The database that includes the emotionally colored texts analyzed in this process is needed so that the following is displayed to a user at the end of processing: 1) texts of processed news with words / word combinations / word forms marked with a certain color, 2) diagrams depicting the presence of specific emotions and their ratio in a particular text and the totality of texts uploaded for a period specified by the user. Word forms can be both words and word combinations, depending on the particular case, which of them is needed in a given sentence.

[0066] The main algorithms used for these actions are as follows: "Unsupervised machine learning" (Unsupervised machine learning), "the FastText embedding", i elastic search (metric classifier and k-NN nearest neighbour method).

[0067] After that, the text obtained after performing the previous step is analyzed by using the rules of syntactic subordination (ensuring cohesion, control, conjunction) with recognition of language constructs within the text and interpretation of selected emotionally colored words / word combinations / word forms. At this step, the procedures for synthesizing the output information and Tables 2, 4, 6 are used. By using said tables, the emotional coloring of the output word forms and their class is finally assessed, and the syntactic structure of the text is recognized. The procedures for synthesizing the output information are required for the final determination of the syntactic structure of the sentence and then of the whole text, and by using Tables 2, 4, 6, the emotional coloring of the syntactic constructs in the word form / sentence / the whole text is calculated.

[0068] The results are visualized in any suitable manner, for example, by labelling the emotionally colored words / word combinations / word forms, and / or by constructing a graph, and / or by constructing a diagram.

[0069] An example of detecting the emotional coloring of an article is given below.

[0070] Reacted to ECOCIDE (45*18): the eco-activist Greta Thunberg met with Zelenskyy in Kyiv on June 29

[0071] President of Ukraine Volodymyr Zelenskyy met with Greta Thunberg. A WELL- KNOWN (70*08) eco-activist arrived today, on June 29, in Kyiv as a member of the International Working Group on ENVIRONMENTAL CONSEQUENCES (29*14) of the WAR (45*18).

[0072] Details Greta Thunberg THANKED (12*24) for the invitation and emphasized that she would engage representatives of environmental public organizations in a dialogue on the main tasks of the International Working Group and drawing attention to ENVIRONMENTAL CONSEQUENCES (29*14) of the WAR (45*18). In his turn, Volodymyr Zelenskyy pointed out that the International Working Group should address a wide range of issues related to the impact of the RUSSIAN AGGRESSION (61*17) on the ecological system of Ukraine.

[0073] Thunberg pointed out that ECOCIDE (45*18) and DESTRUCTION (48*16) of the environment is a form of WAR (45*18), which is known in Russia. She said:

[0074] They INTENTIONALLY (70*14) TARGET (32*14) the environment, the means of living for people, and they also DESTROY LIVES (45*18). After all, this is ultimately about people. This is about UKRAINIANS who SUFFER (16*15) from this WAR (45*18).

[0075] The eco-activist EXPRESSED HOPE (72*26) that various organizations and authorities will be able to collect assessment information about the CONSEQUENCES (29*14) of the DESTRUCTION (48*16) OF THE ENVIRONMENT in order to "CREATE AN OPPORTUNITY (70*95) for BRINGING RUSSIA TO RESPONSIBILITY (47*25) for her ACTIONS AND CRIMES" (34*17), and to CREATE AN OPPORTUNITY (70*95) for the RECOVERY (47*25) of Ukraine "in a persistent way".

[0076] Main emotions:

[0077] Calculating the classes that correspond to emotions. In this text:

[0078] 45*18, fear - 7 (appears 6 times in the text and appears 1 time in the heading of this news);

[0079] 48*16, anxiety - 3 (appears 3 times in the text);

[0080] 29*14, alertness - 3 (appears 3 times in the text); 47*25, approval - 2 (appears 2 times in the text);

[0081] 34*17, indignation - 1 (appears 1 time in the text);

[0082] 32*14, dissatisfaction - 1 (appears 1 time in the text);

[0083] 12*24, friendliness - 1 (appears 1 time in the text);

[0084] 16*15, pity - 1 (appears 1 time in the text);

[0085] So, the conclusion is as follows:

[0086] The main emotions peculiar to the text:

[0087] Fear, with a high degree of manifestation (7);

[0088] Anxiety, also in a high degree (3);

[0089] Alertness (3).

[0090] Additional emotions: approval (2), dissatisfaction (1), indignation (1), friendliness (1), pity (1).

[0091] Thus, the claimed invention provides a method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions / word forms in a text and its intensity, a method for automatically detecting the emotional coloring peculiar to expressions / word forms in a text and its intensity, as well as a system implementing said method, which ensure achieving a technical result consisting in the ability to thoroughly assess the emotional coloring of the text and its intensity, in particular to measure the level of empathy of the text.

Claims

Claims1. A method for training a machine learning model to automatically detect the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includes creating a database of emotions, each of which is assigned a weight coefficient of emotional coloring intensity; creating a database of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; creating a database of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; creating a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; creating a training data set by using the created databases; training a model by using the created training data set to obtain a trained machine learning model.

2. A method for automatically detecting the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includes receiving a text and dividing it into sentences; analyzing the text by using a machine learning model with the labelling of its emotionally colored words / word combinations / word forms with one of the emotions and its weight coefficient; wherein the machine learning model is trained by using a training data set created by using databases: of emotions, each of which is assigned a weight coefficient ofemotional coloring intensity; of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; analyzing the text obtained after performing the previous step by using the rules of syntactic subordination with the recognition of language constructs / word forms within the text and the interpretation of selected emotionally colored words / word combinations / word forms; visualizing the analysis results.

3. The method as claimed in claim 2, wherein the analysis results are visualized by labelling the emotionally colored words / word combinations / word forms, and / or by constructing a graph, and / or by constructing a diagram.

4. The method as claimed in claim 2, wherein at the step of text analysis, by using the machine learning model, a database of emotionally colored words / word combinations / word forms, a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence, and a database of logical and semantic rules of interaction of words / word combinations / word forms in a sentence are additionally used.

5. The method as claimed in claim 2, wherein at the step of text analysis, by using the rules of syntactic subordination, a database of emotions, each of which is assigned the weight coefficient k of emotional coloring intensity, and a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence are additionally used.

6. A system for automatically detecting the emotional coloring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, which includesat least one processor, at least one machine-readable medium communicatively coupled to the at least one processor, and program instructions for detecting the emotional cojoring peculiar to expressions / word forms in a text including multiple words / word forms and its intensity, stored on the at least one machine-readable medium and executed by the at least one processor, which include program instructions for receiving a text and dividing it into sentences; a machine learning model trained by using a training data set created by using databases: of emotions, each of which is assigned a weight coefficient of emotional coloring intensity; of emotionally colored words / word combinations / word forms, each of which is assigned an emotion and its weight coefficient of emotional coloring intensity; of logical and semantic rules of interaction of words / word combinations / word forms in a sentence; of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence and of rules of false bottom; program instructions for analyzing the text by using the machine learning model with the labelling of its emotionally colored words / word combinations / word forms with one of the emotions and its weight coefficient; program instructions for analyzing the text obtained after performing the previous step by using the rules of syntactic subordination with the recognition of language constructs / word forms within the text and the interpretation of selected emotionally colored words / word combinations / word forms; program instructions for visualizing the analysis results.

7. The system as claimed in claim 6, wherein the program instructions for visualizing the analysis results include program instructions for visualizing the analysis results by labelling the emotionally colored words / word combinations / word forms, and / or byconstructing a graph, and / or by constructing a diagram.

8. The system as claimed in claim 6, wherein the program instructions for analyzing the text by using the machine learning model further include instructions for analyzing the text by using a database of emotionally colored words / word combinations / word forms, a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence, and a database of logical and semantic rules of interaction of words / word combinations / word forms in a sentence.

9. The system as claimed in claim 6, wherein the program instructions for analyzing the text by using the rules of syntactic subordination further include instructions for analyzing the text by using a database of rules of working with antonyms of emotions and their effect on the change in the emotional coloring of a sentence.

Images