School in Digital Age: How Big Data Help to Transform the Curriculum

Abstract

The transition to digital society is characterised by the development of new methods and tools for big data processing. New technologies have a substantial impact on the education sector. The article represents the results of applying big data to analyse and transform the learning content of Moscow’s schools. The analysis of the school curriculum comprised the following: (a) identifying one-topic lesson scripts, (b) analysing cross-disciplinary connections between subjects, (c) verifying the compliance of the lesson script digital content to the Federal Educational Standards. The analysed material included 36,644 lesson scripts. The analysis has been conducted using specifically designed digital tools featuring data mining algorithms. The article considers the issue of applying data mining algorithms to analyse school curriculum for the improvement of its quality.

1. Introduction

The transition to digital society is characterized by the development of new methods and tools for big data processing. New technologies have a substantial impact on the education sector. The current transformation of education is characterized by the application of data mining algorithms and big data technologies to the digital learning materials stored in the learning repositories and learning management systems. For the education system of Moscow, the Moscow Electronic School project (MES) was designed and implemented—a digital management system for the city’s school education. By the spring of 2020, the MES offered access to 1,638,275 students, 72,492 educators, and 1,648,362 parents of students of different regions of Russia, including Moscow.

The MES system consists of the electronic diary and the electronic library. The electronic diary is a digital tool that allows teachers, parents and students to obtain such information about the organization of a student’s life at school as the curriculum of each academic subject, lesson schedule, extracurricular activities, schedule of additional classes, the progress and results of educational activities (assessments in subjects, topic, teacher comments), analysis of the dynamics of academic performance in the subject and etc. The electronic library is a digital repository of learning objects.

The MES is a unique repository of learning objects open to every teacher, student, and parent. By May 2020, the MES contained 45,321 electronic study guides, 154,311 applications, 167,227 tests, 1,413,646 lesson scripts, as well as several millions of atomic contents.

The process of creating new digital content by Moscow’s teacher community is continuous and the volume of the uploaded data is increased daily. This is one of the outcomes of the digital transformation of education. Although a contemporary school education is not very different from the school education of the analogue age, it may be assumed that for the first time in the Russian education science, the researchers have obtained access to big datasets of learning content.

During recent years, Moscow City University have been engaged in research using the MES data. The issues of differentiation and segmentation of electronic learning resources collections have been described in the article “The Frontiers of the Moscow Electronic School” (Remorenko and Grinshkun) in 2017 [1]. In 2018–2019, research has been conducted on the causes, explaining high demand for lesson scripts and their quality [2,3], network engagement of teachers with E-learning objects stored in the MES [4], search of data in the MES [5,6], development and design of materials for different subjects in the MES, including materials for special needs students [7,8,9,10,11,12,13], training of pre-service teachers [14,15,16,17,18], and selection and updating of learning content [19,20]. Big data and its impact on the transformation of business processes and structure in the education sphere have been explained in the article “Building a learning culture for the digital world: lessons from Moscow” by A. Schleicher [21]. While Schleicher describes the use of big data in education in general, in our work, we go further by applying text mining methods to analyse the curriculum of school education and show how the analysis of a large sets of data helps to transform the curriculum of school education through solving practical problems that previously could not be solved before new technologies appeared.

Text mining methods are often used to solve information retrieval problems and, in particular, to cluster scientific articles [22]. There are also examples of the use of text mining methods to the curriculum selection development and validation [23,24]. However, these studies are characterized by the use of data sets limited to a small number of subjects or represented by data from only one educational organization or subject area.

Today, many methods of semantic analysis of texts are known. According to [24,25,26], the best results are shown by methods based on neural networks.

The MES data is one of the elements of the post-industrial education as a cultural space that has been defined by I. Levin as “Data Intensive Science”, along with social media and personality online [27,28,29,30]. In the age of ubiquitous access to all kinds of data, it is crucial to conduct research experiments with data and apply new technologies to inform science with larger amounts of data. By using the data “mirror”, we can look into the actual embodiment of subject content that is used for teaching and learning at school, analyse its compliance to the Federal State Educational Standards (FSES), discover the existence of cross-disciplinary connections in the learning content of various school subjects, etc.

The purpose of this research is to study the results and prospects of using big data technologies, in particular text mining methods and approaches to transform the curriculum of school education by the example of solving the practical problem of comparing the curriculum of school education in Moscow with a thematic classifier—the topical framework. The novelty of the research is the use of well-established text mining methods for the study of the curriculum of school education on the city scale. Moreover, this is the first example of applying text mining methods to MES educational data.

2. Materials and Methods

The research focused on the most popular type of digital content in the MES—lesson scripts. A lesson script describes the content and the course of a lesson of any subject in the electronic format. A lesson script may include interactive tasks, schemes, maps, videoclips, tests, etc. Lesson scripts are produced by teachers and used to conduct in-person and online lessons and classes. When designing a lesson script, the author must specify its title, subject, grade level, education level (primary general, basic general, secondary general education), give a short description, create and upload content for all lesson stages on three screens: teacher’s, students’ and common screen. Each lesson script has its own ID.

Since June 2020, every new lesson script is assigned to a topic—a didactic item in the topical framework. The topical framework is a unified structured classifier of learning programmes that is based on the FSES and includes topics and didactic items (elements of learning content). The topical framework was introduced into the MES in May 2020 to solve the issues of structuring learning content and eliminating low-quality learning materials. However, there is still the problem of mapping previously created lesson scripts to the thematic framework. Moreover, as the thematic framework changes dynamically, the problem of regularly comparing new didactic items of the thematic framework to the existing lesson scripts in the MES database remains actual.

A teacher, author of a lesson script, can submit a lesson script for moderation. Moderation is a process of confirming the compliance of a lesson script with the FSES, the current scientific vision, the federal educational legislation, as well as checking for grammar, stylistic or factual mistakes. The lesson scripts that have undergone moderation are uploaded in the MES public space. These lesson scripts are open to all registered users and can be copied, added to favourites, launched or rated.

At the time of the analysis, there were 1,413,646 lesson scripts in the MES e-library, 44,527 of which have undergone moderation. In order to solve the problem of comparing didactic items and titles of lesson scripts the lesson scripts with unique titles have been selected. In total, 36,644 lesson scripts have been selected for the analysis.

The analysis was conducted on the raw data.

Table 1. Example of the data assigned to a lesson script in the analysis.

Script_ID	Subject	Education Level	Title
34105	Environment	Primary general education	Geographic map and plan
34147	Mathematics	Primary general education	Numbers from 10 to 20. Grade 1. Lesson 1.
34214	Environment	Primary general education	Land navigation. Compass
286028	Russian language	Primary general education	Consolidating knowledge on noun cases. Grade 3
60449	Reading of literature	Primary general education	4 gr._V.F.Odoyevsky “A town in a tobacco box”. 4 grade

and

Table 2. Example of the data interconnected in the topical framework in the analysis.

Education Level	Subject	Topic	Didactic Item
Primary general education	Environment	Human and society	I am primary school student
Primary general education	Environment	Human and society	I and my family
Primary general education	Environment	Human and society	Rules of safe conduct
Primary general education	Russian language	Spelling	Hyphenation
Primary general education	Russian language	Spelling	Capital (uppercase) letter at the beginning of a sentence

show the examples of raw data used in the research. Table 1 shows the data describing the lesson scripts. The data include the unique number of the lesson script in the MES, the subject that the lesson script belongs to, the level of education that the lesson script belongs to, and the name of the lesson script. Table 2 shows semantic data related to the thematic framework: the level of education, the subject, the topic and the didactic item.

To compare the lesson scripts and the thematic framework, we had to solve the problem of mapping each lesson script to a relevant didactic item.

Based on the available data, we developed a text mining model and an algorithm that were both aimed at analysing the data and extracting one-topic lesson scripts and didactic items, as well as providing visualization of the results in two-dimensional space.

The text mining process comprised three steps in working with the data:

• Pre-processing of data;
• Defining the proximity of meanings of didactic items and lesson script titles with regard to each other;
• Visualizing the results of comparing didactic items and lesson script titles.

Step 1. Pre-processing of data

The step of data pre-processing included eliminating irrelevant data and making consistent changes in the data display, such as making the letter cases uniform, removing conjunctions and prepositions, removing punctuation marks such as commas, full stops, dashes, etc.

Step 2. Defining the proximity of meanings of didactic items and lesson script titles with regard to each other

At the second step, to define the proximity of meanings of didactic items and lesson script titles with regard to each other, we used the word2vec method [31,32]. At this step, the artificial neural network was trained on a large text corpus consisting of lesson script titles and didactic item titles.

Step 3. Visualizing the results of comparing didactic items and lesson script titles

The third step included visualizing the results of comparing didactic items and lesson script titles in a convenient mode for further evaluation by experts. We decided to visu-alize the data in the form of two-dimensional graphs, where the distance between data points reflected the semantic proximity of lesson scripts and didactic items with regard to lesson scripts. However, at the second step, we obtained word embeddings within the 50-dimensional space. Therefore, for adequate visualization of the data, the number of embedding dimensions had to be reduced while maintaining the distance between the embeddings. To do this, we used the algorithm t-distributed stochastic neighbour embedding (t-SNE) [33].

After the data had been processed by the text mining algorithms, we suggested two hypotheses.

The first hypothesis stated that one-topic lesson scripts must have similar titles, and the algorithm can group them into one-topic semantic clusters. This would allow defining topical correlations among school subjects and topics that generate a large amount of lesson scripts.

The second hypothesis stated that the algorithm can generate semantic clusters of lesson script titles according to their proximity to the titles of didactic items within the topical framework. Thus, we would verify the compliance of lesson scripts with the learning content defined by the FSES.

To test the hypotheses, we conducted the text mining of lesson script titles, and the comparative analysis of lesson titles and didactic items for the subjects of the primary general and basic general education.

The research has its limitations. On the one hand, the research was conducted using only the titles of lesson scripts and didactic items. Other types of semantic data included in the lesson scripts were not analysed. Therefore, we assume that the expansion of semantic data i.e., the inclusion of all content of lesson scripts into the analysed material at the following stages of the research, might alter the obtained results. On the other hand, this research was the first attempt to apply text mining algorithms to solve the task of analysing learning content uploaded in the MES. The important aspect at this stage was to evaluate the performance quality of the algorithm.

The evaluation of the algorithm performance was conducted by expert evaluations. The expert community consisted of the faculty members of Moscow City University with expertise in the subjects of primary general and basic general school levels, including Russian, Literature, English, French, German, Mathematics, Algebra, Geometry, Computer Science, History, Social Studies, Geography, Chemistry, Biology, Physics, Physical Education, Music, Fine Arts, Environment. The experts were asked to assess the titles of lesson scripts and their correlation with the didactic items within the semantic clusters that were generated by the algorithm, as well as to indicate the typical mistakes in the algorithm’s patterns for further improvement of its performance.

3. Results and Discussion

As a result of applying the algorithm, we obtained a vector representation of each lesson script and each didactic item. We also mapped each lesson script to the closest didactic item defined by the cosine similarity:

$$similarity= \cos \left(\theta \right)= \frac{{\sum }_{i=1}^n{A}_i{B}_i}{\sqrt{{\sum }_{i=1}^n{A}_i^2}\sqrt{{\sum }_{i=1}^n{B}_i^2}} ,$$

(1)

where A—vector representation of lesson script, B—vector representation of didactic item, n—set of real numbers.

Figure 1 and Figure 2 show visualization for 36,644 lesson scripts in the MES, based on the vector representation of each lesson script. Every school subject is shown by a different colour. The stars on the graph show the lesson scripts that have been awarded with grants. The Figures show that most of the lesson scripts are clustered by the school subjects that they are connected with. However, a number of lesson scripts were not included in subject clusters which, we assumed, were cross-subject (cross-disciplinary) lesson scripts.

The analysis of the data revealed the following cross-disciplinary topics at the level of primary general education: “The Great Patriotic War”, “The surrounding environment”, “Letters and sounds”, “Regional studies: Moscow”. The basic general level comprised such cross-disciplinary topics as “Plant Classification”, “Poems about Moscow. Marina Tsvetayeva”, “Neurological technologies”, “Healthy lifestyle. Diseases and illnesses”, “People in big cities and their lifestyle”, “Safety basics for everyday activities”, “War”. In the semantic cluster “Safety basics for everyday activities”, there were titles of lesson scripts in the subjects Physical Education, Handicrafts, Computer Science. The semantic cluster “War” included lesson scripts in such subjects as, on the one hand, the world history and history of Russia and, on the other hand, literature and music. These lesson scripts included, for example, “War consequences: revolution and collapse of the empire”, “Russia in the 1st World War”, “The birth of song during war”.

The comparative text mining of the lesson script titles and didactic item titles allowed obtaining the following visualizations (see examples in Figure 3, Figure 4, Figure 5 and Figure 6). The visualizations are based on the vector representation of each lesson script and each didactic item. The blue dots on the graphs denote the lesson script titles, the orange dots denote didactic items in the topical framework by subject.

The visualized data represent the semantic proximity of the lesson script titles and the didactic items in the topical framework by subject and the overall compliance of the lesson script dataset with the FSES, since the topical framework and its didactic items have been developed based on the current educational standards by subject.

Having obtained these results, we proceeded to evaluating the quality of the developed model of semantic data mining. As mentioned above, the quality assurance of the model was conducted by expert evaluation using the dataset of lesson script titles and didactic items in the school subjects of basic general education.

The experts were asked to evaluate the clusters generated by the algorithm using the scale 0/1 for every entry “didactic item—lesson script title”. The table entries contained the didactic item titles and the lesson script titles that had the closest proximity to each other (see example in

Table 3. Example of data mapping representation for expert evaluation.

Didactic Item Title	Lesson Script Title	Expert Evaluation (0/1)
Nuclear energetics	9-grade Physics Atomic nucleus structure
Nuclear energetics	Alternative energetics
Nuclear energetics	Atomic nucleus structure
Barometers	Spotlight 5. Module 6. Extensive Reading. Across the Curriculum: Science
Barometers	“Interconnected vessels”
Brownian motion	Mechanical motion
Brownian motion	Mechanical motion
Brownian motion	Motion

).

We agreed that “0” would mean that the mapping is incorrect, “1” that the mapping is possible. The experts were also asked to describe the reasons of incorrect mapping by the algorithm.

As in the [25,26] studies, accuracy metric is used to measure the performance of lesson scripts to didactic item mapping. It is important to note that, according to the study [25], the accuracy of data mining methods varies from 8% to 77%, depending on the field of knowledge and the algorithm used.

The example statistics of the expert evaluation reflecting the quality of mapping the didactic item titles and the lesson script titles in the basic general education are shown in

Table 4. Results of expert evaluation of the semantic algorithm performance.

N°	Subject	Total N° of Entries	Mapping Evaluation—1	Mapping Evaluation—0	% of Correct Mapping (Accuracy)
1	English language	3041	584	2457	19.20
2	Biology	712	194	518	27.24
3	Geography	1418	680	738	47.95
4	Fine Arts	443	85	358	19.19
5	Computer Science	3075	264	2811	8.58
6	History	849	51	798	6.01
7	Literature	1759	142	1617	8.07
8	Mathematics (incl. Algebra and Geometry)	1573	822	691	56.07
9	Music	263	80	183	30.42
10	German language	519	335	184	64.55
11	Social studies	783	297	486	37.93
12	Russian language	2504	935	1569	37.34
13	Physics	1759	928	684	52.76
14	Physical Education	1262	874	388	69.25
15	French language	289	116	173	40.13
16	Chemistry	230	141	89	61.30
-	Total	20,479	6528	13,744	36.62

. The ratio of correct mapping by the algorithm of the didactic item titles and the lesson script titles varies from 6.01% (History) to 69.25% (Physical Education). These results are consistent with the text mining algorithms quality reported by the other researchers [25,26].

The problem of mapping lesson scripts to thematic items is actually a classification problem, where the didactic item is a class. As researchers [34], we can compare the results obtained with the results of classification, if the lesson scripts were mapped to didactic items based on random selection. This comparison is presented in

Table 5. The comparison of the proposed text mining method with random selection.

N°	Subject	Total N° of Didactic Items (Classes)	Total N° of Lesson Scripts	Random Selection Accuracy, %	Text Mining Mapping Accuracy, %
1	English language	180	3041	16.89	19.20
2	Biology	314	712	2.27	27.24
3	Geography	210	1418	6.75	47.95
4	Fine Arts	88	443	5.03	19.19
5	Computer Science	154	3075	19.96	8.58
6	History	477	849	1.78	6.01
7	Literature	328	1759	5.36	8.07
8	Mathematics (incl. Algebra and Geometry)	568	1573	2.77	56.07
9	Music	82	263	3.2	30.42
10	German language	254	519	2.04	64.55
11	Social studies	240	783	3.26	37.93
12	Russian language	406	2504	6.17	37.34
13	Physics	357	1759	4.93	52.76
14	Physical Education	58	1262	21.76	69.25
15	French language	268	289	1.08	40.13
16	Chemistry	215	230	1.07	61.30
-	Total	4199	20,479	4.88	36.62

.

Table 5 shows that the average accuracy of 36.62 % is significantly (7.5 times) higher than the random selection-based approach. In comparison with the random selection approach, the proposed text mining algorithm shows the best accuracy in Chemistry (57 times higher). Moreover, extremely high accuracy is shown in French language and German language (37 and 31 times higher, respectively). The high efficiency of application of the used method is achieved in disciplines, such as Mathematics, Biology, Social Studies, Physics and Music (20, 12, 11, 10 and 9 times higher, respectively). A good quality of assignment of lesson scripts to didactic items is shown in Geography and Russian language (seven and six times higher, respectively). Acceptable results can also be considered in Fine Arts, History and Physical Education (more than three times higher, then random selection). Low results were achieved in subjects such as English Language and Literature (marginally better than random selection). The worst accuracy of mapping lesson scripts to didactic items is shown in Computer Science. As it is worse than the random selection-based approach, a separate study in this discipline is required.

According to the expert evaluation results, the most typical mistakes of the algorithm included the clustering of lesson script titles into semantic clusters based on:

-

similar word, one-rooted words, interference with prepositions;

-

similar lemmas;

-

close meaning of words;

-

proper names;

-

generalized definitions;

-

prepositions;

-

word abbreviations;

-

arithmetic operation symbols;

-

Roman numbers;

-

overlapping of letter sets.

The algorithm also makes mistakes in clustering lesson script titles when they incorporate several meanings or ambivalent wordings. An example of this may be the lesson script, with the title of a famous Russian fable “The Dragonfly and the Ant”.

The large number of mistakes made by the algorithm shows that the algorithm lacks sufficient “knowledge” to cluster incomplete titles, and that lexical topics are mapped to the grammar code, which is not applicable.

Besides, it was impossible to map the lesson script titles in foreign languages to the didactic item titles in Russian. This is why the visualizations of French language and German language subjects show that the didactic items are located far from the lesson scripts (see Figure 4).

Within the course of the research, we discovered that the distance between the objects (titles of didactic items and lesson scripts) inside the semantic clusters depends on the structure of the lesson script titles.

Here is an example of the semantic cluster in the subject English language, generated by the algorithm with regard to the didactic item “Healthy Lifestyle” (Figure 7 and Figure 8).

Area 1 of this semantic cluster includes lesson scripts with titles “Health is more important than wealth”. “Healthy lifestyle. Grade 6”, “Grade 7. Healthy lifestyle. Control lesson (two lesson scripts)”. Area 2 includes lesson scripts with titles “Healthy lifestyle. Food”, “Healthy lifestyle. Healthy nutrition”, “Grade 7. Healthy lifestyle. Conclusions”, “Grade 5. Healthy lifestyle. Conclusions”. Area 3 includes lesson scripts with titles “Healthy lifestyle. Food-5”, “Healthy lifestyle. Food-6”, “Healthy lifestyle. Food-3”, “Healthy lifestyle. Sports exercises”. The distance between the objects depends on the semantic emphasis in the topic “Healthy Lifestyle”. In this example, we see that all the names of lesson scripts containing the words “Healthy lifestyle” are combined by the algorithm into a separate cluster and the distance of the names of lesson scripts from didactic items depends on the composition of words in the names of lesson scripts, which demonstrates the correct operation of the developed algorithm.

Another example shows the semantic cluster of Physical Education. Figure 9 represents the semantic cluster “Sports games components for physical games”.

Area 1 of this semantic cluster comprises such lesson scripts titles as “Components of Basketball. Active sports game. Five strikes”, “Basketball components for physical games”, “Football components for physical games”. Area 2 includes lesson scripts with titles “Basketball components for physical games. Grade 2 (3 lessons)”, “Basketball components for physical games. Grade 3”, “Basketball components for physical games”. Area 3 comprises lesson scripts with titles that include the phrase “Active games with volleyball elements”. Area 4 comprises lesson scripts with titles that include the phrases “Pitching components for physical games”, “Track-and-field athletics components for physical games”. Area 4 comprises lesson scripts with titles that include the phrase “Active games with components of pitching”. Area 5 comprises such lesson scripts titles as “Outdoor games with elements of football”. It should be noted that since the model has been developed by means of training the neural network, the quality of algorithm performance in terms of clustering one-topic lesson scripts can be further improved by ensuring that it is trained on larger text corpora.

We have obtained these results for all subjects. However, there have also been some “peculiar” results of mapping didactic item titles to lesson script titles. For example, in the subject History, the semantic clusters were formed not around one didactic item, but around semantic clusters containing didactic items and lesson script titles.

Here is an example of a semantic cluster that has been generated by the algorithm (see Figure 10) with regard to the didactic items of the semantic cluster “Rebellions”.

The didactic items and lesson scripts in this cluster include the word “rebellion” in their titles. The didactic items comprise “Pugachev’s Rebellion”, “Rebellion on Don lead by K. Bulavin, Jacquerie”, “Rebellion of W. Tyler”, “Astrakhan Rebellion”, “Bolotnikov’s Rebellion”, “Decembrist rebellion”, “Rebellion in Bashkiria”, “Rebellion of Stepan Razin”, “Spartak’s Rebellion”. We can see here that the algorithm includes one-topic semantic cluster didactic items with identical words. This semantic cluster also includes (without line connections) lesson scripts with the titles “Decembrist rebellion. Grade 8”, “Decembrist rebellion on December 14, 1825”, “Decembrist rebellion (consolidating learned material and conclusions)”, “Spartak’s rebellion (three lesson scripts)”.

Figure 11 shows the semantic cluster generated by the algorithm with didactic items that incorrectly include different historic persons. For example, M. Loris-Melikov, Zh. Garibaldi, Voltaire, Miltiades, V. Kornilov, P. Milyukov, and M. Skobelev have been grouped into a one-topic cluster.

The semantic analysis has also revealed that titles of certain didactic items are so broad that neither algorithms nor experts are able to correctly assign lesson script titles to didactic items. This is the case when didactic item titles do not relate to concepts but are worded as, for example, “Relationship of music and art”.

4. Conclusions

The developed algorithm for semantic data mining and visual representation of the data allowed analysing the structure of lesson script topics in the MES learning repository. Large semantic clusters unite topics that are popular among teachers in terms of creating lesson scripts, for example: lesson scripts in “ Addition and subtraction” (Mathematics, PGE); lesson scripts on topics related to tenses of the verb (English language, PGE); lesson scripts on general topics such as “Spotlight”, “Starlight”, “City” (English, BGE); lesson scripts on the topic “Track-and-filed athletics” (Physical education, BGE); lesson scripts on the topic “Punctuation marks” (Russian language, BGE).

In the course of the research, we have discovered popular cross-disciplinary topics such as “War”, “Healthy lifestyle”, “Rules of safe conduct”, “City”, etc. However, the structure of lesson script titles in the common semantic field shows that school subjects tend to be semantically isolated from each other.

The semantic analysis of the titles of lesson scripts of the basic general education subjects revealed the topical closeness of lesson script titles, as well as the clustering of similar lesson scripts with titles that begin with words that do not always have semantic significancy for understanding the meaning or topic of a lesson script. For example, the gene- rated semantic clusters included one-topic lesson scripts, as well as lesson scripts with titles that begin similarly or have common words. We also noted the clustering of lesson script topics by common meanings or similar topics. Nevertheless, there are semantic clusters that have been generated due to technical peculiarities of data display.

The comparison of lesson script titles and didactic item titles visualized the comp- liance of lesson scripts with the FSES (in the semantic aspect). At the current stage of the research, the accuracy of the results produced by the text mining algorithm is quite low, however, in case of further training and development of the algorithm, we will be able to receive instant visualization, showing compliance of learning content to the FSES.

Therefore, it is essential to train the algorithm: create a list of stop-words, keywords, as well as other methods that can boost the algorithm’s performance.

The analysis of the semantic mapping results shows that it is necessary to improve the topical framework. The topical framework must be structured based on the ontology of concepts, not just items of learning content.

Besides, the application of text mining methods to the learning content contained in the MES e-library will allow more advanced development (including requirement specification) of the topical framework for the whole school curriculum, and transformation of approaches to structuring learning content using big ideas. The results of the research can be applied to structure the learning content of different digital learning systems in school education. This will streamline integration of learning content from other systems into the MES or the exchange of learning content between different digital systems in school education which, in the long run, will improve the quality of learning content, facilitate the indication of various types of content and accelerate resources with high teaching potential, as well as raise the overall quality of learning materials.

The analysis of the MES data clearly shows that the application of new methods and approaches to big data processing, as well as using artificial intelligence technologies, can change our views on education, even today. With the help of these technologies, we can discover new unconventional knowledge about the course and content of teaching and learning processes, which leads to the changes in the educational system and reflects the transition to digital society.

Further research is planned to improve the quality of comparison of lesson scripts to didactic items by:

-

expanding the semantic data set by including annotations and full texts of lesson scripts in the analysis;

-

translating topics and didactic units of the thematic framework into foreign languages (Chinese, English, German, French) for semantic analysis of the names of lesson scripts in foreign languages.

In the future, it is assumed that the developed algorithm will be able to automatically map educational content uploaded to the MES with the thematic framework, determine the “best” version of the lesson script among similar ones (most accurately revealing the topic), and discard low-quality content.