Computerized content analysis of articles from the journal "Bulletin of Finance, Industry, and Trade" for the year 1917: testing the capabilities of the artificial intelligence module in the MAXQDA program.

Voronkova Dar'ya Sergeevna Postgraduate student, Section of Historical Informatics, History Department, Lomonosov Moscow State University 119192, Russia, g. Moscow, ul. Lomonosovskii Prosp., 27 korp.4, kab. G-432 dasevo@yandex.ru Other publications by this author

In the era of widespread interest in data Science and the constant growth of data volume, artificial intelligence is actually becoming a means not only of processing data, but also of analyzing it. It can be used to solve a wide range of research tasks, including expanding the potential for meaningful interpretation of results.

The need of modern humanitarian research for computerized tools for working with texts has led to the emergence of programs for mixed (qualitative and quantitative) methods of analyzing and visualizing text documents.

Computerized content analysis (content analysis) has become a popular method of text analysis over the past few decades. The most reputable software for its implementation is MAXQDA, developed by Verbi. The program has the ability to perform all stages of content analysis: text markup (creating a system of semantic categories and indicators), determining the frequency of occurrence of these categories, including joint ones, which allows you to find out their interrelationships, visualization of results (including network analysis),

Artificial intelligence tools (AI Assist module) have now been developed and included in the program as an assistant for working with large collections of texts, which are the subject of research in our article. Our study and testing of the capabilities of the artificial intelligence module within the framework of MAXQDA will allow us to show the new potential of modern content analysis, which successfully reveals a layer of implicit (hidden) information in the data.

The purpose of the AI Assist module is to deepen the research of sources. The task of the developers was not to replace humans with artificial intelligence, but to provide the user with a set of tools that can make his work easier.

This article discusses the results of testing artificial intelligence tools in a computerized analysis of articles in the journal Bulletin of Finance, Industry and Trade for 1917. The journal contains valuable information on such areas of economics as finance (including government strategy in various aspects of financial policy), taxes, and international trade. Reform proposals and polemics on these issues are also found on its pages.

In the revolutionary year of 1917, all the materials of Vestnik ... reflected the process of changes in the domestic (and not only) economy, the historical background of which was the fourth year of the First World War.

Computerized text analysis using special software as a research method does not lose its relevance ^[1], has long been successfully used, including for the study of the historical press. ^{[2, 3, 4, 5]}. This type of source stores valuable information, which is becoming more and more laborious to extract using traditional methods, and in some cases not so effective. The Bulletin of Finance, Industry and Trade has already been the object of application of modern computerized content analysis ^[6]. At the same time, the journal has not previously been investigated using the potential of artificial intelligence. The purpose of this article is to analyze the capabilities of the AI Assist tool in annotating and paraphrasing text, indexing and encoding it, as well as in obtaining reference and auxiliary information in chats with a virtual assistant.

Source characteristics of the "Bulletin..."

The magazine was the official publication of the Ministry of Finance of the Russian Empire. There are different points of view regarding the year of the beginning of its publication – 1883, 1884 or 1885. In 1865-1884, the "Index of Government Orders for the Ministry of Finance" was published, the "heir" of which was the "Bulletin ...". Continuity was expressed even in the figure of the editor-in-chief, since in the process of a rather large-scale reorganization of the publication, which followed in 1885, A. K. retained this position. Veselovsky (previously the head of the "Index ..."). It was after the reorganization that the editorial office became independent (albeit formally, being a government mouthpiece, but with its own staff and budget). The circumstances that led to the appearance of the journal in the context of the tax reforms of the 1880s were discussed by the author in the already mentioned article, where the hypothesis of 1884 as the initial year of publication of Vestnik was also put forward and substantiated. The biographies of the chief editors in 1917, as well as the author's composition of this year's journal and the subject matter of the articles were analyzed earlier ^[7].

Research methodology

The MAXQDA software was selected for the implementation of computerized text analysis ^[8]. It was in its version MAXQDA 24.5.1 that the capabilities of an artificial intelligence tool, the AI Assist module, were implemented for the first time. Artificial intelligence has become one of the central themes of MAXDAYS 2024, the annual MAXQDA virtual user conference held on September 10-11, 2024, which was attended by the author of this article. The main report by Dr. Stefan Redicker was titled "Integrating AI Into Qualitative Data Analysis with MAXQDA: Methodological and Practical Questions (and Answers)". The materials of the MAXDAYS 2024 methodological seminars (Spotlight Sessions) are already available on the official website ^[9], and the videos are posted on the MAXQDA YouTube channel ^[10].

Key AI Assist features at the moment

Let's first consider the AI Assist functions for generalizing uncoded data (i.e., data that is not assigned any categories; the latter are called codes in MAXQDA). The results of the development of the code system for the "Bulletin ..." for 1917 have already been published ^[6]. The functions that allow you to deepen the analysis of the source include:

· Summary of the document

· Summary of the highlighted text

· paraphrasing the selected text fragment

· summary of the paraphrase

Take, for example, the tenth issue of Vestnik..., published on March 12 (25), 1917. It published articles "From the Editorial Board" ^[11], "A new work on the theory of money" ^[12], "Legislation on enterprises of enemy subjects" ^[13], "Russian-American economic relations" ^[14], "Foreign trade of Russia in 1916" ^[15]. Despite the variety of topics in the articles (including the most important one about the editorial board's attitude to the February Revolution), the AI Assist module focuses on the structure and dynamics of Russian exports. The summary results are saved as a document note. We have made two attempts: in the first case, a longer summary in the form of a bulleted list, in the second case, the average volume of the summary is set in the form of solid text (Fig. 1; hereafter, some screen forms of the MAXQDA program have been edited to improve their readability).

Fig. 1. Fragment of the summary of issue 10 of the "Bulletin ..." for 1917 using AI Assist

In the longer version, there is a typo for a whole century – 2016 instead of 1916. But in general, the text can be considered quite "meaningful". The focus on foreign trade is probably due to the fact that the "trade" category is indeed one of the main ones in this issue. Apparently, AI Assist focuses on the volume of text occupied by a topic. However, if you look at Figure 2, the question arises: why did the assistant ignore the "finance" category, if it occupies the first place in number 10, and "trade" is in the second?

Fig. 2. Diagram of coded segments in issue 10 of the "Bulletin ..." for 1917 (built in MAXQDA by the author)

To demonstrate the result of the summary of the highlighted text, let's turn to the already mentioned program article "From the Editorial Board". It expresses the attitude towards the accomplished February Revolution. Figure 3 shows how AI Assist understands the meaning of this article.

Fig. 3. AI Assist summarized the article "From the Editorial Board" in issue 10 for 1917.

Given that this function does not include a critical analysis of the text (the virtual assistant, like ourselves, of course, cannot know how objective the editorial board was in this case), the content is reproduced fairly accurately.

* * *

It may be necessary to rephrase the text of the source so that the main idea of a fragment becomes clearer to the reader. Paraphrasing in text analysis programs is used for several purposes: it helps to make the text more understandable and accessible to the reader, especially if the source document contains complex or special terminology; helps in highlighting the main topics for analyzing large volumes of text; helps in translating texts into other languages, preserving meaning and context. Paraphrasing can also be useful for understanding the information presented in diagrams, figures, and tables. Finally, paraphrasing in a scientific article can serve as an alternative to direct citation. Note that, unlike annotation, paraphrasing changes the wording in the text without adding additional information and comments to the source text, its purpose is to improve the readability of the text.

The volume of AI Assist's "response" when paraphrased in the beta version is limited to 511 characters. Using the example of almost the entire first paragraph of the article "Increasing taxation in Italy" ^[16], published in the first issue of Vestnik ... in 1917 (our choice is due to the fact that the search for additional state revenues was so relevant in the conditions of the First World War that the authors of the magazine often turned to foreign experience), let's see, how artificial intelligence paraphrases taking into account the limitation (Fig. 4).

Fig. 4. A fragment of the source text and its paraphrasing using AI Assist (in the frame on the right)

AI Assist was not mistaken: the very large figure of Italy's military spending – more than a billion lira per month – did not escape its attention. The main reasons for this, which are mentioned in the text, are listed, as are the measures planned by the government to find sources of financing for military needs. Moreover, it took 266 characters to do this, which is even less than the limit.

In version MAXQDA 24.6 ^[17] (released on October 22, 2024), it became possible to summarize all paraphrased text fragments and save them as a document note (Fig. 5-7).

Fig. 5. Framed - the result of the execution of the menu item

"To summarize the paraphrases" of the "AI Assist" clause

Fig. 6. Settings for summarizing paraphrased text fragments

In addition to choosing the language of paraphrasing, there are settings for the length of the summary text (short, medium, long), displaying the result in a bulleted list, and an option for additional instructions, giving the researcher even more options.

Fig. 7. Summary of paraphrased text fragments of the first issue of Vestnik...

It can be seen that AI Assist is quite applicable in cases where it is necessary to summarize the text and highlight the main thing in its content.

* * *

You can ask AI Assist to clarify if the meaning of terms is unknown or in doubt (such queries are stored as notes inside documents). This feature will especially help, given that historical sources may contain both outdated and highly specialized vocabulary, which the researcher is not always familiar with. In this case, it is not so important whether a code is assigned to a text fragment or not, in fact, the option does not generalize, but explains the data. Although the translation of its name into Russian is not entirely correct (Fig. 8).

Figure 8. The option is called "Explain the choice of text".

It would be more correct to call it "Explain the highlighted text"

Let's try to find out from AI Assist what the "metal theory" is ^[18]. Figure 9 shows that the explanation is generally detailed, while the assistant assumes that the user is familiar with the concept of "fiat money" (i.e. money not backed by gold or other precious metals – their value is set and guaranteed only by the state).

Fig. 9. AI Assist explains the concept of "metal theory"

The entire process of computerized data analysis in MAXQDA is based mainly on assigning codes to this data, which are actually analytical categories. The coding possibilities are wide, it is carried out using various program tools.

* * *

Both during the initial construction of the category-code system and during its further adjustment, AI Assist's ability to offer new codes and subcodes will be in demand. Let's test these features with an example. In the beta version, AI Assist cannot encode text longer than 5,000 characters. Therefore, we will select a small fragment of the obituary article "Paul Leroy Beaulieu (1843-1916)", dedicated to the memory of the famous French economist, corresponding member of the St. Petersburg Academy of Sciences: "Paul Leroy Beaulieu, forty-four years ago, contributed to the establishment of the journal "French Economist", which he was the editor-in-chief throughout this period and in in which he gave an editorial every week (with one single exception). The issue dated Saturday, December 9, 1916, also contains his article on the war, while the author of this article died on the night of Friday to Saturday from malignant pneumonia" ^[19]. It would be possible to encode this passage in the "obituary" code. And here's how AI Assist did it (Fig. 10).

Fig. 10. Codes proposed by AI Assist for a text fragment

Let's note the pros and cons. The first include two variants of thematic codes (AI Assist offers them based on "understanding" the content of the article) and interpretive codes (what can we say about the article as a whole?); the fact that the codes will not be applied to the text automatically, but only if the user chooses to do so; saving comments of artificial intelligence in the form of notes to the codes. The disadvantages include the fact that AI Assist actually only retells the source text – there is a lack of some detachment and academic style. The assistant is so "impressed" that among his suggestions there are such codes as "Dedication", "Tragic fate", "Symbolism of the publication date", "Dedication to memory". On the other hand, let's think about whether it is appropriate to mark "an article about the war published shortly before the author's death" with the code "Timeliness of publications"? At the same time, the phrase "article on war" occurs three times in AI Assist – he "understood" the subject of the mentioned publication by Paul Leroy Beaulieu well.

In 1917, the press continued to discuss the prospects of the government's policy of resettlement of peasants, in particular, to the regions of Siberia. We will ask AI Assist to offer subcodes of the "resettlement policy" code (this code contains the least encoded segments – 156; the limitations of the beta version will not prevent us). We can see what happened in Fig. 11.

Fig. 11. Subcodes of the "resettlement policy" code proposed by AI Assist

The result is quite satisfactory here. We have configured its display as a list. The list is extensive (6 items), adequately reflects the essence of the source code. If you need an in-depth analysis of this particular topic, the AI Assist suggestions will help you with your work.

* * *

Let's move on to working with encoded text and see how AI Assist is able to generalize encoded data.:

· Summary of coded segments;

· Code summary;

· summaries in summary tables (summary tables in MAXQDA allow you to see all encoded segments of any code in documents at the same time - this is in demand, because it immediately gives the overall picture, thus AI Assist makes it easier to work with large amounts of data).

For example, the code segments "mail and telegraph" (194 of them) are summarized as follows (Fig. 12).

Fig. 12. AI Assist compiled a summary of the coded segments of the "mail and telegraph" code

And again, the result is acceptable: there is a mention of the reason for the growth of departmental revenues in 1915 and a number of planned further measures.

Another example. Let's say we know that the most frequent code in issue 43 of Vestnik ... for 1917, the very issue that became the last for the magazine as a whole, is the code "political sphere". We established this by relying not on AI Assist, but on other MAXQDA tools (in particular, the code cloud). By the way, both codes and documents can be activated all or selectively during analysis. How AI Assist summarizes this code in only one document is shown in Fig. 13.

Fig. 13. AI Assist summarized the code "political sphere" in issue 43 of Vestnik... for 1917.

It may seem surprising why government savings banks have fallen into the category of "political sphere" rather than "finance". However, this is the result of auto-encoding with a dictionary, which we performed earlier based on frequency tables of words and then truncating them to the base. Figure 14 shows the dictionary for the code "political sphere".

Fig. 14. Dictionary for the code "political sphere"

It seems justified that the basis of "States-" falls into this category. The vocabulary in MAXQDA is built in such a way that indicators cannot belong to more than one category at the same time.

In order to create a summary of the pivot table using AI Assist, we previously used the Code Explorer tool (its main function is to review the occurrence of code in our project in MAXQDA) and found that the most frequent in the "Bulletin ..." for 1917, the code "finance" is most often (399 times) found in 42 the issue of the magazine. Then we activated both the code and the document, created a pivot table, and used the resume option using artificial intelligence inside it. The result is shown in Fig. 15.

Fig. 15. AI summary for the pivot table (fragment)

Issue 42 contains the article "Experience in analyzing and solving a financial problem (graduation)" ^[20], which really focuses on the prospects for military loans. In particular, its author proposes to introduce a compulsory loan, while AI Assist "did not notice" the involuntary nature of this measure.

* * *

One of the main innovations of AI Assist is the ability to ask him questions in chat mode, and without leaving the MAXQDA 24 workspace. The historical events reflected on the pages of Vestnik ... are separated from us by a long time distance – more than a century. Not all the realities of this crucial historical moment are clear to the reader. And there are not always available sources to consult in case of doubt. Finding the right information can take a lot of effort. Therefore, there is no doubt that chats will be useful.:

· with a document,

· with encoded segments of a specific code.

Figures 16 and 17 show the windows that appear on the user's screen when each of these functions is launched.

Fig. 16. AI Assist chat start window with a document

Fig. 17. AI Assist chat launch window with encoded segments of the selected code

Obviously, a chat with a document can be used for thematic analysis and consideration of various opinions on some significant issues.

The chat with encoded segments also has prospects for further development and refinement of the MAXQDA code system (categories) by the researcher, which will also make the analysis more in-depth and reveal the implicit content of the source.

Unfortunately, there is currently no data on which large language models (LLM) developers have trained AI Assist on. This may be the subject of discussion outside the scope of this article. The assistant can produce results not only in the language of the source document (therefore, it would be useful to know which data arrays and in which languages were used in the training of the module).

Let's take a closer look at how chats work in AI Assist. For the chat with the document, we chose the first issue of the "Bulletin ..." for 1917. First, we asked the virtual assistant to list the topics covered in it (Fig. 18).

Fig. 18. AI Assist, in chat mode with a document, lists the main topics of the first issue of Vestnik... for 1917.

AI Assist correctly reflected the subject matter of the first issue of the magazine in 1917. We know that it contained articles "Theory and practice of monetary circulation", "The next tasks of the state forestry administration after the war" ^[21] and "Raising taxation in Italy".

And why, as we have just seen in Fig. 18, are playing cards so important to the Italian government that a state monopoly is imposed on their production? Let's ask AI Assist about this (Fig. 19).

Fig. 19. AI Assist finds and lists the reasons for the introduction of a state monopoly on playing cards in Italy

It turns out that the playing cards produced in Italy were exported. The monopoly affected the foreign market, and the prices of cards within the country were regulated. Such measures also provided the state with additional income during the war.

We will test the chat with encoded segments using the example of segments of the "neutral" subcode included in the "country" code. This code was created to analyze the reflection in the "Bulletin ..." of the participation of the warring parties in the First World War. The code includes subcodes "Entente and Allies", "Triple Alliance, etc." and "neutral". Of course, it would be interesting to pay attention to the warring blocs, but, unfortunately, for now AI Assist processes texts with less than 100,000 characters in chats with encoded segments, and the volume of mentions about the parties to the conflict is greater.

So, "neutral". In what context are we talking about such countries? The AI Assist response is shown in Fig. 20.

Fig. 20. The context of the mention of neutral countries in the "Bulletin ..." for 1917, according to AI Assist

It turns out that the main neutral states of Europe on the pages of Vestnik in 1917 were Sweden, Switzerland and the Netherlands. And in Latin America, Argentina is one of them. And what about Norway? Let's try to find out (Fig. 21).

Fig. 21. Norway in the "Bulletin ..." for 1917

Norway was first mentioned as a country trading with Persia in 1915-1916. It is becoming a new transit route to the East. Following the example of Denmark and Sweden, Norway introduces a tax on military income.

Let AI Assist suggest to us how neutral countries can be differentiated (Fig. 22).

Figure 22. AI Assist offers criteria for differentiating neutral countries

Quite correctly. The criteria identified by AI Assist are the degree of economic dependence of neutral countries on belligerents; political orientation (full or partial neutrality); degree of participation in military supplies to opponents; demographic characteristics; level of economic development. Each criterion has subsections.

* * *

Many researchers have been waiting for the appearance of the text auto-encoding function using artificial intelligence. And now AI Assist has learned how to do it. The process in our example is structured as follows: we selected the code "war" and activated the number 42 "Bulletin..." in the document system (we first made sure through the already mentioned Code Explorer that such an important code for 1917 is most often found in this issue – 163 times). A window opened in the code note (Fig. 23), in which AI Assist sets the encoding criterion (you can specify the exclusion conditions, but we are focusing only on searching for segments where the war is definitely mentioned).

Fig. 23. Coding criteria for AI Coding (beta). It is noticeable that not all areas of the window are translated into Russian.

Then AI Assist performed auto-coding according to the specified criterion. The result is stored in the code system, provided with an explanatory comment and marked as a code created by artificial intelligence. In the future, it will be possible to combine it with the existing "war" code, study it in more detail, delete it, etc. To quote one of the fragments encoded by AI Assist: "The war and especially the revolution catastrophically changed the city budget. Not to mention the extraordinary expenses for wartime needs (to cover them with loans will be discussed below), but ordinary city expenses for the maintenance of existing urban institutions and enterprises, due to the high cost of all goods in general and due to the huge increase in salaries of employees and workers, have increased enormously" (Fig. 24) ^[22].

Fig. 24. The result of AI Coding

Undoubtedly, the AI Coding function has a great future. On the one hand, it has become possible to use artificial intelligence support in automatic encoding mode, and on the other, the user has full control over this process, since the whole point of this tool is that auto-encoding is based on predefined descriptive criteria, which the assistant must demonstrate in each case. And this is the difference between AI Coding and auto-coding with a dictionary: MAXQDA also has this feature, but there the focus is on the occurrence of only certain word forms in the text.

MAXQDA Tailwind (Beta version)

While the article was being prepared for publication, we could not ignore the release of the new version of the MAXQDA program – 24.7.0, released on December 17, 2024 ^[23]. And on January 30, 2025, the author received an invitation to become a beta tester of the new AI Assist functionality – MAXQDA Tailwind ^[24].

The developers set ambitious goals for themselves – this follows from the fact that the new toolkit, although included in the AI Assist module, received its own branded name MAXQDA Tailwind ("tailwind" means "tailwind"). The early access period for beta testers of AI Assist Free users was limited to only seven days from the moment of activation, with no possibility of extension, so it is especially important to reflect this experience in this article. The beta version of MAXQDA Tailwind has the following main features:

· projects (no more than 5) and importing data into them (no more than 20 documents each, and the document size should not exceed 20 KB and 120,000 characters);

· Automatic summary of each document;

· Thematic analysis: artificial intelligence highlights the topics covered in all documents;

· Summary tables.

Let's look at them in more detail. The interface includes a sidebar and a main workspace. The tabs in the sidebar (which can be minimized) change depending on whether we are in the window of all created projects or inside one of them.

You cannot create a project without a description (Fig. 25). On the one hand, this is a kind of "intrusion" into the researcher's activity, and on the other, it helps to organize and structure it.

Fig. 25. Project name, description, and source language settings in MAXQDA Tailwind

As you can see, MAXQDA Tailwind is trained to work with documents in 10 languages (except Russian: English, German, Spanish, Turkish, French, Italian, Portuguese, Japanese and Chinese). We'll talk about working with the project below. In the meantime, let's pay attention to the "Projects" button: these are all projects. The "Chat" button has the inscription "Coming soon" next to it – you will be able to use it later. The Feedback button allows the beta tester to keep in touch with the developers.

The "Help & Support" button (Fig. 26) will give access to the user's manual, which is currently being updated, help you contact customer support, find answers to frequently asked questions, learn about the protection of data processed by artificial intelligence, and also familiarize you with legal information and privacy regulations.

Fig. 26. The "Help & Support" button in MAXQDA Tailwind

The project window actually consists of a folder with a description and a separate button for creating a new project, if necessary (Fig. 27):

Fig. 27. Project window in MAXQDA Tailwind

When we move into the project, there is, of course, no data in it yet, which we are informed about, indicating the acceptable formats (.docx, .rtf, .txt) and the already mentioned file size limits.

After importing the MAXQDA data, Tailwind starts analyzing it (Fig. 28):

Fig. 28. MAXQDA Tailwind analyzes files in a project

The author of the article decided to check how the "assistant" would cope with the analysis of 20 documents (the numbers of the "Bulletin ..." in Russian in modern spelling). Unfortunately, the limit at the beta stage turned out to be an insurmountable obstacle for artificial intelligence - the picture you see in Figure 29 did not change for more than two hours, and the process had to be interrupted without waiting for the result. Therefore, our further research will be based on the material of only one, the tenth issue of Vestnik... for 1917.

The result of the automatic summary of this number is shown in Fig. 29.

Fig. 29. MAXQDA Tailwind summarized the number 10 of the "Bulletin ..." for 1917.

Compared to the summaries of the same issue, which we have shown in Fig. 1, there is a step forward: now the article "From the Editorial Board", expressing its attitude to the February Revolution, has not been ignored by artificial intelligence. Although, since the entire issue is summarized, attention to this particular article is not as close as when using the "summary of selected text" function (Fig. 3). Obviously, each approach has its pros and cons. As we remember, in the summary of a document using AI Assist (and not specifically MAXQDA Tailwind), you can select the length of the text – shorter or longer – and also customize the display with solid text or a bulleted list. MAXQDA Tailwind, at least in the beta version, lacks flexible settings, but it allows you to immediately go to the paragraphs on which the resume is based (their numbers are highlighted in green and active). The disadvantage is that the number is actually perceived by MAXQDA Tailwind as one big article. Artificial intelligence is not "surprised" that at first we are talking about "significant changes in the Russian political system and the transition to freedom of speech," in the very next paragraph – "about Professor M.I. Tugan-Baranovsky's book "Paper Money and Metal." In fact, in this case, we have a summary of the second article number 10 – "A new work on the theory of money" ^[12].

* * *

Let's move on to the thematic analysis using MAXQDA Tailwind.

Figures 30-31 show which 5 topics were found.:

Fig. 30. Proposed MAXQDA Tailwind based on the analysis of the topic document (beginning)

Fig. 31. Proposed MAXQDA Tailwind based on the analysis of the topic document (end)

Artificial intelligence has identified the following topics: "Theory and practice of monetary circulation in Russia", "State regulation of the economy during the war", "Foreign economic relations of Russia: dynamics and prospects", "Influence of the political situation on economic thought and practice", "Economic recovery of Russia after the war: problems and prospects". Each of the topics is accompanied by a brief description. Those topics that meet the research objectives can be checked for further work with them (which in MAXQDA Tailwind is placed in the "Topics" tab of the project sidebar). Interestingly, there are only 5 topics offered, but in the bottom line of the window we see the inscription "Topics in project 0/20", i.e. each project can have up to 20 topics, including those manually added on the "Topics" tab. Here we again find active references to the context on the basis of which artificial intelligence has identified a particular topic. It is possible to reformulate the description of topics, add a topic, or perform a thematic search again, as well as export the results in the format.xlsx. Let's repeat: all project documents can be thematically analyzed, it's simple (including due to the limitations of the beta version of MAXQDA Tailwind), we show this functionality using the example of one document - number 10 of the Bulletin...

* * *

And the last thing we will test in this article is pivot tables in MAXQDA Tailwind. Pivot tables allow you to see all encoded segments of any code in documents at the same time (this makes it easier to work with large amounts of data).

To create a pivot table, you need to select at least one file (the only one in our test project) and one topic. The result (fig. 32) can be exported in the same format as the themes file.

Fig. 32. Fragment of the pivot table inside the project in MAXQDA Tailwind

The user can search through the pivot tables, edit their names, and access the source document by clicking on the paragraph numbers highlighted in green.

Conclusion

Using the method of computerized content analysis, using the latest artificial intelligence tools, it was possible to reveal the information implicitly contained in the journal Bulletin of Finance, Industry and Trade. Thus, as we have seen, artificial intelligence can significantly not only facilitate, but also improve the work with a historical source and make its analysis more in-depth.

The article tested all currently available functions of the AI Assist virtual assistant built into the MAXQDA program: generalization of uncoded and encoded data, explanation of terms, chats with documents and code segments, auto-coding according to user criteria, as well as the beta version of MAXQDA Tailwind. In the latter case, it is especially noticeable how much importance developers attach to the development of the artificial intelligence module: the product, being part of AI Assist, actually represents a new concept of working with data (even if only text): organizing them into projects, tabs inside each of them, thematic analysis (including using pivot tables), fast access to the source documents.

Let's not be too critical of those features of AI Assist that can be assessed as disadvantages. These are, for example, ignoring some codes when summarizing a document, typos, and incorrect translation of options into Russian (or lack thereof, as in the case of code notation during AI Coding). Basing AI Assist on the source text may also not always provide the results that researchers need. We saw this using the example of the codes proposed by artificial intelligence for a fragment of an article about the death of Paul Leroy Beaulieu. In addition, we hope that developers will soon remove restrictions on the amount of text being processed, in particular, when paraphrasing, offering codes, and in chats with encoded segments. In general, at the moment it is noticeable that the smaller the text AI Assist analyzes, the better it does it.

Speaking about the positive experience of working with AI Assist, we note the adequacy of summaries of selected text fragments, paraphrasing, and explanations of terms. Breakthrough technologies have also been implemented in their own way: in chats, AI Assist shows itself almost like a real assistant, rather than offering some abstract maxims on query topics. In AI Coding, the situation is generally not bad either – the assistant follows the criteria quite carefully, explaining the results of his actions.

The artificial intelligence tools in MAXQDA 24, which are included in the additional AI Assist module, are being actively upgraded. It is possible that at the time of publication of the article, new features will be added to them (given that the work on the text began by the author of the article at the beta stage of AI Assist, and ended with the beta testing of MAXQDA Tailwind). Anyway, all of them are promising and useful for both beginners and experienced users.