This paper presents the results of interdisciplinary project that is a combination of computer program and psycholinguistic approach to media study. In the research we presented the programs that can be used for monitoring and analysis of media content to identify hate speech at its early stage. The aims of research were the following: 1) develop content analysis program for monitoring Russian media outlets; 2) apply the psycholinguistic approach for identifying hidden and manipulative hate speech. In the research there were used two types of content-analysis: quantitative and qualitative. Quantitative content analysis was conducted with computer program that was developed to select publication that could have contained hate speech. For qualitative content analysis the psycholinguistic method of text analysis was used. The method applies for identification methods and tolls that journalists use to incriminate hidden and manipulative hate speech. It is hypothesized that programs of content-analysis help to optimize work and makes it less time-consuming and more efective for analyst, journalists and other specialists who involved into media study. Methods. Quantitative content analysis, psycholinguistic method of qualitative content-analysis. Quantitative content analysis was developed with Python programming language. The publications were selected according to the key words, periods of search (month) and the name of outlet. The list of key words includes words that are used in media for discrimination, dehumanization, and marginalization of objects of hate. Implementation such a program helped to reduce time of monitoring of media outlets. The qualitative content-analysis was conducted with the authors' psycholinguistic method of text analysis that can be applied for analyzing media texts. The programs of content analysis were applied within the project “Hate Speech in Online Media Publicizing Events in Crimea”. The results were published in a data analysis report on spreading the hate speech in the Russian language media communicating the armed Ukraine - Russia conflict and events related to it in Crimea on a regular base (December 2020 - May 2021). The research showed that the content analysis programs used in the project are useful tools for systematizing and processing data in humanities research and can be used by a wide range of specialist who have deal with collection and processing of information (media, communication, human rights and so on).
Nowadays, people deal with large amounts of information from diferent online resources. According to
the Digital 2020 Global Overview, the average person spends 6 hours and 43 minutes online per day, or
approximately 40% of their time [
Examining large amounts of textual data is a time-consuming and labour-intensive process that can be
simplified by using computer programs that allow quicker and more eficient processing of information
and receive reliable quantitative results [
At the same time, if information analysis is part of a professional activity, a specialist needs additional special skills for monitoring, processing, and analysing data. For example, a such specialist as media analyst, PR, journalist etc. has not only to collect and analyse information, but also to monitor sources to find a certain type of information. Therefore, it would be advisable to introduce special courses in the curriculum to teach the basics of big data analysis for future humanities professionals whose work involves processing and analysing information. In this context, teaching content analysis skills allows reducing time and conduct research, proceed data and analysis information more efectively. For example, for civil activists such skills allow efectively identify texts that contain manipulation, disinformation and hate speech. In our study, we are talking about journalists, human right defenders, and other specialties related to media analysis and processing.
In the context of our work, we will consider an example of the content analysis software application for conducting quantitative and qualitative research of the online media space. The study was conducted as part of a project involved a technical specialist, a psycholinguist, the human right activists, and journalists. The task of the project was to monitor Russian media to observe the situation before the military aggression, because we consider media as an important tool for understanding the vectors of strategic communication in the country. The results of the study (2014–2021) revealed that the Russian media actively used covert and manipulative hate speech aimed at creating a dehumanised, marginalised, and demonised image of Ukrainians as speakers of the Ukrainian language and culture. Hidden and manipulative hate speech was a signal that should have drawn public attention to further discrimination and genocide against the targets of hate.
As we see, the importance of monitoring hate speech at its initial stage is stated in documents of
international organisations and research by scholars. For example, the organization “Anti-Defamation
League” use “Pyramid of hate” that illustrates the prevalence of bias, hate and oppression in society and
demonstrates the progressive intensification of hate: from supporting each next step to genocide. The
lowest level “bias attitude” includes stereotypes, insensitive remarks, microaggression and so on. The
upper levels “bias motivated violence” and “genocide” include act of violence and annihilate an entire
people [
In the 2017 annual report of the Council of Europe highlights the danger of the dissemination and
amplification of poor-quality information in media that is divided into three types: mis-, dis-, and
mal-information, which are difered based on the dimensions of harm and falseness: mis-information is
when false information is shared, but no harm is meant; dis-information is when false information is
knowingly shared to cause harm, and mal-information is when genuine information is shared to cause
harm. The report stressed that hate speech is used to cause harm and discriminate against a group of
people on religion, race, and other grounds: “. . . people are often targeted because of their personal
history or afiliations. While the information can sometimes be based on reality (for example targeting
someone based on their religion) the information is being used strategically to cause harm” [
In recent years, there has been considerable interest in issues of the negative influence of mass media in humanities (psychology, linguistics, sociology, etc.), IT, and Computational linguistics.
In linguistics, much work on the potential of speech propaganda and manipulation has been carried
out by Bulyigina and Shmelev [
There is a vast amount of literature on specific features of media representations of war and military
conflicts (Pocheptsov [
In computational linguistics, the study and detection of hate speech explore by using natural language
processing. For example, Schmidt and Wiegand [
The aims: 1) develop content analysis program for monitoring Russian media outlets; 2) apply the psycholinguistic approach for identifying hidden and manipulative hate speech.
Content analysis is a quantitative and qualitative method of analysing content (text, pictures etc.)
in order to identify and measure various facts and trends reflected in these materials. The purpose
of quantitative content-analysis is to fins valid resources and to collect certain data The purpose of
qualitative content analysis is to organize and elicit meaning from the collected data and to draw
realistic conclusions from it. The utilization of content analysis in Humanities and Social Sciences
are described by Dey [
The peculiarity of content analysis of a media text is its direct correlation with external circumstances, such as the socio-political situation. Content analysis is based on counting the occurrence of determined components in the analysed information materials, supplemented by identifying statistical relationships and analysing the structural links between them.
The need to use content analysis for analysis of large text arrays was caused by the development of mass communications in the late nineteenth and early twentieth centuries. The content-analysis was used for analytical research of texts in the media outlets. This method can be used as the main method, for example, to analyse the political orientation of a publication, or as an auxiliary or control method with other methods, for example, to measure media efectiveness. Content-analysis can also be applied alongside with other methods, for example linguistic and psycholinguistic, for evaluating the efectiveness of a media outlet.
One of the main founders of the research procedure is the sociologist Lasswell [
The classical model of content analysis proposed by Lasswell [
Lasswell’s research became the basis for the development of further procedures for analysis of media
texts and answering a variety of questions, ranging from the peculiarities of the worldview of the
average citizen to the ideological orientation of a particular media organ [
Today, various types of content analysis are used both for analysing large data sets and for analysing a single text. Because of a large demand of such programs. IT industry is actively developing software to improve functionality and customise content-analysis programs to analyse big text data, which parses text from unformatted content and unstructured data from social media, news reports, surveys, etc. to provide practical information such as the mentioning frequency of a certain brand, personality event, or counting particular words in the texts (e.g., programs for semantic content analysis such as Semantrum, Nvivo, MAXQDA, Yoshikoder, Advego, SentiStrength, Nvivo, etc.). The brief characteristics of some of these programmes are given below.
ADVEGO is a free programme for semantic analysis of texts. It performs text statistics by the number
of words, determines the most frequently used words and their number, as well as words and phrases
that are part of the semantic core [
Yoshikoder is a free programme that allows you to find online texts by a given phrase or expression.
This programme allows you to examine keywords-in-context, and perform basic content analyses, in
any language. At the same time, in our research we need to use some specific keywords that are not
included in the dictionary [
Programme SentiStrength which refers to the programme for Sentiment Analysis or Opinion Mining.
The programme is set up to search for evaluative and emotive vocabulary in the text according to a
pre-compiled dictionary, which is divided into groups: “negative” and “positive” vocabulary, which is
placed on the scale of emotional intensity of the statement [
MaxQDA, NVivo software can be applied in a range of sectors: from social science and education to
healthcare and business. These programs can be used to analyse data from interviews, surveys, field
notes, web pages, and journal articles. These programmes are similar in terms of functionality and
operation [
In the research the following methods were used: quantitative content analysis, psycholinguistic method of qualitative content-analysis. Quantitative content analysis was developed with Python programming language. The publications were selected according to the key words, periods of search (month) and the name of outlet. The list of key words includes words that are used in media for discrimination, dehumanization, and marginalisation of objects of hate. Implementation such a program helped to reduce time of monitoring of media outlets. The qualitative content-analysis was conducted with the psycholinguistic method of text analysis that is the authors’ method that can be applied for analysing media texts.
The study outlines the approaches and opportunities for interdisciplinary interaction of humanities and
computer sciences. The research of hate speech was done with collaboration with the Crimean Human
Rights Group (Siedova and Krylova-Grek [
II. Quantitative content analysis. This stage involves 1) monitoring of the selected media outlets; 2) identification of groups in relation to which negative characteristics are applied; identification of keywords used to create a negative image of the selected groups; 4) texts selection. III. Qualitative content analysis. At this stage, we applied the author’s psycholinguistic analysis of the selected texts.
For conducting quantitative content-analysis, a special program was developed by analyst and technical specialists of the Crimean Human Rights Group O. Sedov (now the program is under the process of patent). The program allows to optimize the teamwork by surveying a large array of information to find a pool of texts that can incite xenophobia and hatred.
The program sets up the following parameters for hate speech content analysis: 1. Online resource name. We specified the names of news agencies that specialize in current news.
The initial cohort includes the nine most popular online media, whose trafic ranges from 1 million to 15 million visitors monthly. 2. Keywords. We entered words that usually accompany texts with hate speech. To single out keywords, we developed Hate dictionary based on the careful analysis of publications in selected media. Monitoring and word selection took place in 2017-2018. The dictionary includes more than 400 words. It should be noted the dictionary is constantly supplemented and changed due to the emergence of new narratives, concepts, and words. 3. Search period. We set the time interval: date and year. In our research, the most appropriate search period that can provide us with the required amount of data is one month. For example, from 1 of August to 31 of August.
The content analysis program searched the selected websites to find in which hate speech regarding such ethnic and social groups might be used: Ukrainians, Crimean Tatars, Jews, Residents of Crimea and Donbas Russians, Activists and Journalists, Euromaidan Participants, LGBTQ groups. Most of the hate words referred to Ukrainians.
Key words included new created lexicon, archetypes of the second world war and other words that marginalized some ethnic and social group. Studies found that articles featuring hate speech often use terms and concepts that are not recorded in the oficial dictionaries of the Russian language. Some of these words are ‘made up’, created purposefully to incite hatred. In some cases, such words are formed by combining parts of words denoting nationality and obscene lexical units, for example, “kriptobanderovtsy” (a compound word made up of “crypto” and “Banderite”), “natsgady” (a compound word made up of a shortened form of Nazi/Nationalists that sounds similar and is used by Russian media as a play on words; and the word ‘gady’ that is similar to ‘foul people’ in English (skunk, despicable), “khokhlodauny” (a compound word made up of ‘khokhly’ (see above) and the Russian word for Down’s syndrome suferers ([douny] /n., pl.(ukr.)/).
Apart from other things, we considered a well-known vocabulary available in glossaries, and with negative connotations, that was used by journalists in the investigated media in relation to objects of hatred, ridicule, marginalisation, and so on. Most such words go back to of WWII archetypes, and common negative stereotypes. For example, “fascist”, “fascism”, “nazi”. Moreover, based on the phonetic similarity of the words “Nazi” and “nationalism” in Ukrainian and Russian, journalists use the word [natsist] instead of [natsionalist] (Nazi, nationalist).
Among other words are those that humiliate and marginalise another people’s language by distorting the phonetic sound of words for sarcastic or mocking reasons, and bracketing the phonetic spelling of words (the transliteration of Ukrainian words into Russian in terms of our study), which contextually, in the publications, is sarcastic and expresses contempt for the Ukrainian language and its speakers (e.g., “svidomyye”, “nezalezna”).
The study materials were monitored and sampled using a content analysis program: texts were selected using set key words and word combinations. Such words and word combinations had been collected by the project’s monitoring group during preliminary studies into the hate content of Russianlanguage online media and public social networks. Nine Russian-language sites were searched. The Russian language vocabulary used in relation to the main ethnic groups, as well as to the most vulnerable social groups of the population, who currently live in the territory occupied by the Russian Federation, was added to the list of key words and word combinations.
As an example, we delve into the interface of the hate speech content analysis program in greater detail. First and foremost, it’s important to highlight the distinctive features of the program’s terminology: “query words” refer to the specific words we input into the search based on our Dictionary of Hate. On the other hand, “keywords” are the words that most frequently appear in the text of a publication. For instance, let’s take the online platform Politnavigator as an example. We selected the keywords for September 2019 (figure 1).
The program’s interface presented us with the following information: • the total number of articles containing the selected keywords (both in their titles and within the text); • a list of these articles along with corresponding links; • a list of the keywords identified within the articles; • the frequency of each keyword’s appearance within the articles; • the number of query words used.
Upon hovering over a specific article, the interface provided the subsequent details: • a direct link to the respective article; • a hyperlink directing to the media outlet’s website where the article is published; • the keywords featured within the article; • the frequency and number of the keywords; • the frequency and number of the query words.
This interface design allows for a thorough examination of hate speech content, aiding in the analysis and understanding of its prevalence and distribution across various publications.
The content analysis program was used to electronically process the content of the selected websites for the indicated period regarding abovementioned key words and word combinations. The sample of publications produced by the content analysis program was subject to a psycholinguistic analysis exercise. The content was divided into two groups based on the exercise results: publications featuring hate speech, and publications with other types of manipulations.
The texts selected through content analysis underwent psycholinguistic analysis to distinguish between those containing hate speech and those that did not. The content analysis program selected all texts that contain key words, but not all the texts consists hate speech. The reasons are related to the algorithm for configuring the content analysis programme which scans the page together with comments and other information. Therefore, the reasons for the errors are justified by the following factors: 1) comments under the text included hate vocabulary. As the research aimed to analyse the products of the media specialists’ activities, the comments were not taken into account and such texts were attributed to error. Among other things, comments can be a product distributed by bots or specifically hired people, which requires additional technical methods for their analysis; 2) texts in which keywords have a direct meaning; for example, the word “fascists” used in the text give a factual retrospective to the military events of the Second World War. At the same time, there were only a few such texts (2%) that were removed from the list for further analysis.
Subsequently, the texts containing hate speech were scrutinized with psycholinguistic analysis to
identify the methods and techniques employed by journalists in these publications [
After having conducted the textual analysis, hate speech was divided into three types, which are characterized by the specific linguistic and graphic tools used in the publication: Type 1: direct hate speech; Type 2: indirect (hidden) hate speech; Type 3: manipulative hate speech.
The monitoring period of online media are from December 1, 2020, to May 31, 2021. We think that
and the amount of content for this period is enough to validate the study outcomes. The project
monitoring group obtained 1,284 publications when keyword-based electronic content sampling had
been completed. The content was processed according to the psycholinguistic method of media text
analysis, and the result was that 560 publications featuring hate speech elements selected from the
entire content were received by the project analytical experts. These include 16 texts with Hate Speech
Type One; 341 texts with Hate Speech Type Two; and 203 texts with Hate Speech Type Three [
Type 2. Indirect use of hate speech is characterised by: marginalisation of the other party with usage of ofensive ethnonyms, polarisation, dividing a society into in-group and out-group, generalisation (attributing one case to an entire group), negative sarcasm and irony, the use of archetypes and stereotypes that develop a certain world view and attitude, the creation of new concepts with negative connotations. Type 3. Manipulative hate speech is characterised by: substituting the meanings of concepts that create negative associations; using fake; use the opinion of biased or ‘pseudo-experts’ citing (people who have little or no experience or knowledge about the problem on which they comment); distorting and misinterpreting historical facts; justifying aggression or violence against a target group;enhancing informative messages with non-linguistic means (photograph, pictures); using manipulative titles that does not match, or distorts, the information presented in the text of the article.
The study investigated eleven Russian language online media, including: five news websites, with
news items on the situation in Crimea largely dominating the content, the audiences of over 1 million
readers per month, and a minimum 25% share of Ukrainian readers; three Russian news websites
regularly writing about Crimea and Donbas, with audiences of over 1 million readers per month, and a
minimum 25% share of Ukrainian readers; two news websites, with largely dominating content that
describes the situation in Crimea, financed out of the Russian Federation budget; and the oficial website
of the “government” of the Pravitel’stvo Kryma (figure 2). The analysis of the auditoriums at selected
sites was conducted using SimilarWeb, a platform that ofers insights into global digital trafic [
The reason for including texts that did not contain hate speech in the search results is as follows: a) certain articles did contain words from the dictionary, but they corresponded to their direct meaning, for example, the word “down” was used to refer to people with Down syndrome; b) the program counted words and expressions contained in the comments under the publication. The second aspect was the main reason for the large percentage of inappropriate texts.
The efectiveness of using a quantitative content analysis programme to monitor hate speech is determined by the following criteria: 1) the amount of time a specialist spends on selecting texts: a specialist spends up to 15 minutes on selecting articles; 2) search criteria: in the software, you can set the search period, the source to be analysed, and a fairly large number of keywords that can be entered, for example, in our study we identified about 400 words related to hate speech;
As a result, reducing the time spent on quantitative analysis allows you to spend more time on qualitative analysis of texts. Another advantage of the quantitative-qualitative approach is that, unlike other search engines and programs, it has better functionality for selecting texts, reduces the time for text selection, and when combined with psycholinguistic analysis of texts, allows us to go beyond simple mathematical calculations and analyse the psycholinguistic methods and techniques used by journalists to influence the minds of the audience.
We would like to express our special thanks of gratitude to Oleksandr Sedov for creating the quantitative content analysis program and the civil organization “The Crimean Human Rights Group” for selecting and providing research materials.