Overcoming cognitive overload of students through the design and development of the structure of the e-learning system

Bragina Elena Vladimirovna Scientific Adviser, Designer of Learning Environments 199034, Russia, St. Petersburg, ul. Embankment Makarova, 6 nedoeduru@yandex.ru Other publications by this author

1.                  Introduction

In the pedagogical context, the modern period is called “the transition from offline learning to online learning” ^{[2, p. 60]}. The rapid development of e–learning systems (hereinafter referred to as SEA) in the last 20 years has radically changed the way educational resources are presented and delivered to students "out of space and time", ensuring greater accessibility, adaptability and efficiency of educational resources.

E-learning is defined in the pedagogical literature as the use of information and communication technologies to improve the quality of education by providing access to resources and services, as well as remote exchange and cooperation using interactive dialog communication networks ^{[1; 10]}. E-learning thus provides a flexible approach to learning and is seen as an educational process that allows you to transfer knowledge and skills to a large number of recipients at different times and in different places. Massive open online courses (MOOCs), SEA developed by various educational institutions, social networking sites, forums and other advanced technical platforms and online platforms turn e-learning from computer to interactive ^[19]. Today, every major university has an e-learning portal for its students and teachers. The main purpose of the introduction of e-learning in the educational process is to increase the adaptability, productivity and effectiveness of full-time learning.

Despite the fact that the problem of e-learning has become widely known, becoming the subject of acute scientific discourse, not all its aspects have been fully studied ^{[30; 34]}. The development of SEA and educational technologies in the last 10 years has shifted the attention of researchers to the problems of independent learning. The unsolved problems include the design and development of SEA that meet the needs of students and the level of technology development. Many researchers note the impossibility of using traditional pedagogical solutions and mechanistic transfers of offline methods to e-learning in e-learning. ^{[2; 4; 5; 6; 7]}. As F.L. Greitzer notes, like traditional classroom teaching, e-learning is largely based on behaviorist paradigms of computer learning, therefore, as a rule, they reflect the philosophy of passive learning ^{[18, p. 2065]}. In addition, as the number of students and the volume of data increases, the quality of e-learning services decreases and the cognitive load in systems increases ^[24]. Fine-tuning and personalization of knowledge or services in SEA should be provided to each individual student in a manner appropriate to his needs. However, the task of building a well-organized, cognitively effective and at the same time providing adequate support to students of SEA still remains unresolved ^{[13; 24; 26; 33; 35]}. Despite the fact that a lot of research has been devoted to improving the cognitive effectiveness of e-learning, little is known about the development of SEA that meets both computer presentation and the cognitive needs of the user.

This article presents some results of a study of the problem of developing a pedagogical model for expanding the context of adult learning in the higher education system, conducted by the author in 2021-2023. The purpose of the article is to substantiate certain theoretical and practical aspects of the model in terms of the design and development of the SEA structure. Among the tasks set by the author is the analysis of modern e-learning systems in terms of their types, functions, architecture and problems. For this purpose, as well as for the development of individual elements of the SEA structure, a systematic review of the literature was conducted. We searched three databases, including PsycINFO, Google Scholar, and e-library. Our search was not limited to the field of education, as the theory of SEA, cognitive load and cognitive schemes was discussed in several fields, including psychology and social sciences. Search keywords included "e-education", "e-learning system", "schema", "schema-based approach", "schema-based learning" and "schema-based learning design", "knowledge map", "knowledge map-based learning". The initial search revealed 3,784 studies in Russian and English. We have excluded duplicates to include only the most recent versions of research and conference materials. Then we excluded studies that were not published in academic publications such as magazines or newspapers. As a result, a source base of 318 studies was formed. The bibliography of this article, due to objectively existing restrictions to the list of sources and literature presented by the journal, consists of 39 most relevant sources.

2.                  Research results

2.1. Basic e-learning systems

E-learning systems facilitate the planning, management and delivery of content for e-learning. Based on the target users and cost, existing e-learning systems can be classified into two main types: mass open online course platforms (MOOCs) and learning management systems (LMS).

MOOCs are online courses open to an unlimited number of participants that are offered by many universities and institutions on the Internet. The term was first used in 2008 by Canadian researchers Dave Cormier and Brian Alexander ^[16]. In the fall of 2011, Stanford University offered the first MOOC course, which was initially registered by more than 160,000 students from around the world, and eventually about 20,000 of them took it. Then three significant MOOCs on the Udacity platform.

There are two types of MOOCs based on different learning theories: sMOCS (connectivist MOOCs) and hMOCS (extended MOOCs). Currently, the most popular and influential MOOC vendors are Coursera, edX and Udacity.

LMS are e—learning systems for hosting, assigning, managing, reporting and evaluating e-learning courses. Many higher education institutions in Russia use LMS as the most important educational tools ^[5] to support course management and improve interaction between students, teachers and information resources, as well as to identify gaps in learning, implementing a wide range of pedagogical methods to promote the educational process.

2.2. Architecture of modern e-learning systems

Many traditional frameworks have been used to create and improve the effectiveness of e-learning systems. One of them is based on service-oriented architectures (service-oriented architecture, SOA) ^[10], which make it easy to expand the educational and functional capabilities of the system by dynamically adding services. The architecture provides, firstly, the independence of e-learning systems, mobile applications and external applications, as well as provides reliable data exchange and interaction between them, and, secondly, the possibility of an innovative object-oriented architecture for implementing e-learning systems on a single software platform to meet the requirements of various e-learning scenarios. Abstract data objects (from the English ActiveX Data Objects — "ActiveX data objects"), encapsulating private memory together with some methods, are widely used as the main components of functional objects, such as courses, announcements, curriculum, etc. This architecture is highly modular, since documents and objects can be created independently, as well as reused using a flexible nesting or containment mechanism ^[11].

The availability of high-speed networks, low-cost computers and data storage devices has led to significant progress in cloud computing technology, which is the use of a network of remote servers hosted on the Internet on request to store, manage, and process data, rather than on one or more local servers. Computing resources are made more accessible and shared by a cloud platform that gathers heterogeneous and distributed devices into a common pool ^[26].

The general structure of SEA is presented in the work of Liu M. and Yu D ^[26] (see Fig. 1), which consists of three logical levels that contribute to improving the effectiveness of teaching and learning:

1) data presentation level;

2) the level of the e-learning system;

3) database level.

2.2.1. Presentation level

The presentation layer focuses on human-computer interaction, providing end users with an accessible user interface and learning resources. This level is aimed at improving the usability, accessibility, reliability and usability of learning ecosystems. Therefore, it is necessary to use proper interface methods, such as HTML, XHTML, CSS and JavaScript, to support mobile learning, in which the presented pages can be displayed correctly in the browser to meet the requirements of device compatibility.

Fig. 1. General structure of e-learning systems

Source: ^[26]

2.2.2. The level of the e-learning system

The level of the e-learning system is aimed at synthesizing educational resources through various functions, such as course enrollment and management, user profile and actions, evaluation and feedback of teaching or learning, communication with the user or cooperation, and so on. It can also be the integration of related components that support a learning model or a learning model ^[27]. For most MOOCs and LMS, this level plays an important role between the presentation level and the database level, and is also a teaching and learning platform that allows each student to flexibly access certain educational resources.

2.2.3. Database Level

At the database level, data created with the help of e-learning systems is placed. This is where training data is collected, stored and used. Because of individual differences, collecting massive data and preserving diversity and dynamic characteristics is very important. In addition, all collected data must be stored until they are used. Typically, existing e-learning storage solutions rely mainly on a relational database such as Mysql and Oracle. The Moodle database is usually MySQL or Postgres, and can also be Microsoft SQL Server or Oracle. Sakai and Blackboard can also be deployed in SQL or Oracle environments.

In addition, NoSQL databases are increasingly being used for large sets of distributed data due to a flexible and scalable architecture. So, MongoDB has chosen Open edX for storing large files such as text files, PDF files, audio/video clips, etc.

2.3. The main problems of the development of e-learning systems

Despite the advantages that e-learning offers, there are still a number of pedagogical and technical problems that need to be solved ^{[13; 24; 26; 33; 35]}. One of the main problems is that educational interaction is largely dominated by “traditional” models controlled by the teacher, which leave independent students with few opportunities to integrate their personal interests and knowledge into the learning process ^[35]. The most important structuring resource for e-learning is still various educational (electronic) manuals. The problem lies not in the teaching materials themselves, but in the role that the text acquires in teaching; students of online courses, when testing knowledge by testing, are often required to reproduce the content of lectures in one form or another ^[33] instead of using this content for comprehension. On the other hand, due to the impossibility of adequate information and pedagogical support, the success of training in SEA largely depends on strengthening the external motivation of students, which is fixed by the assessment system on the course. Students often have a distorted idea of "learning": a significant number of students consider learning to be synonymous with exam preparation ^{[21; 22]}. It is impossible to teach some practical skills at all in SEA, so that students cannot fully understand and assimilate certain elements of the course content ^[13].

As some researchers rightly point out, modern SEA is mostly focused on resources, not on the cognitive process ^{[24; 33]} and focuses on the organization of knowledge resources ^[26], and not on the discovery of hidden learning objects that students need. Due to the rapid growth of e-learning resources, students are increasingly suffering from cognitive overload and perceive learning gaps as an insurmountable problem ^{[36; 38]}. One of the main problems of SEA, such as MOOCs and LMS, is the use of new pedagogical and cognitive approaches to achieve effective transfer and delivery of SEA resources ^[18]. High cognitive load leads to the fact that students assimilate the content of the course to an insignificant extent ^{[4; 20]}. However, most of the literature we have analyzed on e-learning focuses specifically on retention rates or the quality of content. In order to solve the problems associated with the features of e-learning, the design and development of SEA should focus on human cognition, engagement, flow and motivation to promote the benefits associated with learning, such as self-control and a sense of achievement ^{[20; 22; 39]}.

2.4. Theory of cognitive load in e-learning

There are three main research questions concerning the theory of cognitive load and SEA ^[19]: 1) how to reduce the load on working memory in SEA; 2) how to stimulate the consolidation of new information in cognitive circuits in long-term memory through the design and development of SEA; 3) how to measure cognitive load for the purposes of designing and developing SEA. Cognitive load theory ^[24] effectively copes with the limitations caused by working memory by creating instructions that reduce the internal, external and relevant (based on information consolidation) cognitive load on working memory ^{[14; 29]}. At the same time, the goal of researchers in the field of cognitive load theory and SEA is to develop methods for managing cognitive load caused by an educational task in order to increase the success of e-learning. An effective e-learning environment can integrate most of the principles of cognitive load theory relatively easily ^[21], providing multitasking and nonlinear information organization. To do this, electronic SEA should be designed in such a way “to cover the main components associated with the circuit, including circuit hierarchization, circuit construction, circuit automation, circuit activation and/or interaction with the circuit" ^{[22, p. 271]}. The process of designing and developing SEA based on cognitive schemes proposed by I. Jan and co-authors in a recent meta-review of the literature is presented in Figure 2.

Fig. 2. General framework for designing e-learning systems

Source: ^[22]

A schema is a cognitive structure that helps organize and process incoming information ^[9]. The scheme allows a person to distinguish the key features of an object from other non-essential information; that is, the previous knowledge and beliefs of the student play a key role in the construction of more complex cognitive structures. The newly obtained information is combined with existing and easily accessible knowledge in the process of building a scheme. Students develop their own scheme, including elements from low-level schemes into higher-level schemes ^[8]. The theory of cognitive schemas asserts that students' prior knowledge helps them to carry out in-depth cognitive processing ^[37]. Theoretically, using the principles of cognitive schemas can help students create, automate, refine and modify schemas, thereby creating the desired learning process applicable in all fields and contexts.

Another approach designed to reduce the cognitive load on students' working memory is to allow them to learn from a variety of examples ^[13]. “When students develop a consistent and well-structured scheme by studying numerous examples, they can overcome cognitive dissonance caused by new information” ^{[14, p. 1180]}. A well-developed scheme, created on the basis of familiarization with numerous examples, extensive practice and educational support, allows the student to use minimal working memory in the future.

Learning using visualized structures also contributes to the unloading of working memory and the development of a cognitive scheme for students in relation to the concept. It was recognized ^{[14; 15]} that the visual structures of objects help students pay attention to the key features of the objects they learn about. Visual structures combined with hints or additional actions of students are even more effective in terms of acquiring knowledge. In this regard, knowledge maps (hereinafter referred to as KZ), as a tool for their visualization, are used to display the relationships between learning sources and knowledge ^[6]. They are representations with nodal connections, in which ideas are located in nodes and are connected to other, related ideas through a series of labeled links ^[17]. KZ have an advantage when it comes to simplifying relational complexity, and they can be considered as an important cognitive strategy and resource for creativity, problem solving ^[32]. In addition, maps are useful for learning knowledge: with the help of a knowledge map, a student recognizes important points of knowledge and the relationship between them, as well as systematize knowledge. Therefore, they are widely used to solve a wide range of problems in education and the design of educational environments.

With the development of visual computer systems, the importance and complexity of knowledge visualization for SEA design are increasing. Electronic learning materials based on KZ can help students understand various relationships and knowledge structures, improve academic performance and attitude to learning ^[7]. In Figure 2, we see that knowledge mapping is a stage of analysis in the overall process of designing SEA based on schemes. The KZ here is a navigation tool that shows the “flow” of knowledge and guides the learning process ^[17]. This can help students concentrate on the learning process and transform implicit knowledge into explicit knowledge ^[31].

It should be noted here that the number of educational resources available on the Internet is growing dramatically, however, the growth of resources does not automatically imply the growth of students' knowledge. The reason for this can be summarized in the form of three problems of many electronic systems that relate to their architecture: 1) the problem of acquiring knowledge; 2) the problem of knowledge representation and 3) the problem of knowledge resource management. Thus, continuing the analysis of the problems of modern SEA in this section, it should be pointed out that modern SEA is mainly focused on the organization of knowledge resources, and not on the discovery of “hidden from their eyes learning objects” that are in demand by students. Modern SEA mainly generalize knowledge within the framework of resource detailing ^[13], represent knowledge in a hierarchical structure or ontology. This problem can be illustrated by the example of search in Google Scholar and Wikipedia, which are used for the purposes of independent e-learning, including for the search for academic publications. Google Scholar provides users with links to publications according to their relevance to the keywords entered. Publication associations are represented using the "group" and "cited" functions. To determine what they really need, users often have to look through the list one by one, which is undoubtedly time-consuming. Wikipedia is an online encyclopedia consisting of articles organized into hierarchical categories; related articles are linked or cross—referenced using highlighted text. His organizational idea is based on taxonomic categories of knowledge, and not on the cognitive mode of learning. In addition, the "article" is the only detail available, so when students only need to study individual information in an article, such as a definition or algorithm, further manual search is often required. When solving the problem of organizing resources in e-learning, it is necessary to answer the following questions: how can I get knowledge from resources without manually searching inside the resource, taking into account the degree of detail of knowledge available to students? how to present knowledge in such a way as to reduce cognitive overload, taking into account student associations? how to access and effectively manage knowledge resources? Since the "resources" are too large and the "concepts" are too small to meet the cognitive needs of students, it is necessary to offer a more relevant detail of knowledge.

2.5. On the way to efficiency: overcoming cognitive overload through the design and development of the SEA structure

To overcome the problem of a resource-oriented approach to the design and development of SEA, it is necessary to develop a framework for the representation of knowledge in SEA, including a hierarchical scheme and ontology.

As we indicated above, the hierarchical scheme is widely applied to e-learning resources, but the contour does not meet the cognitive needs of students, since they are unable to create associations within knowledge. In web2.0, an ontology is introduced to represent knowledge through concepts and their relationships ^[3], which is consistent with the cognitive pattern of students. However, the main disadvantage of ontology is that the concepts in ontology are not a direct need of students, since it cannot provide any detailed information. In addition, most traditional knowledge resource management systems are based on the Java client-server platform, which makes it inapplicable for large-scale online learning systems.

The design of SEA can be based on an ontology-based way of organizing resources and navigation based on knowledge units and knowledge maps. Having studied the key technologies in SEA, presented in the database of sources analyzed by us, it can be argued that it is necessary to develop:

? tools for displaying an extended knowledge map;

? methods of extracting knowledge units;

? establish associations of knowledge units;

? methods of merging knowledge maps;

? Develop a cloud platform infrastructure for e-learning.

In this article, we will focus on how knowledge can be effectively acquired and presented in SEA based on knowledge maps, the structure of which is being designed by the author at the moment.

2.5.1 Framework

Instead of providing a traditional hierarchical scheme of content in SEA, we suggest using the KZ to facilitate navigation (Figure 3 shows the SEA framework). In contrast to the static content contour, SEA can guide students through the inherent links of knowledge both at the level of concept and at the level of knowledge units. In order to ensure interaction with SEA, it is necessary to allow students to upload resources for the joint creation of KZ. If the local knowledge maps created by students are verified by administrators, these KZ can be combined into a global knowledge map.

Fig. 3. The SEA Framework

2.5.2. Presentation of knowledge

In the SEA being developed, knowledge maps will be used to represent knowledge that provides students with multi-level navigation and collaborative interaction. Appropriate applications are being developed to meet these requirements:

• navigation on the knowledge map: This application provides students with a QA-based interface that allows them to navigate through the internal associations of knowledge of the subject area and accurately determine their learning goal using a three-step approach "Concept" - "Unit of Knowledge" - "Resource";

• Collaborative knowledge mapping: Global knowledge maps can be enriched by allowing users to create knowledge maps based on local resources. However, creating annotations manually based on annotations would be significantly time-consuming and inefficient. To solve this problem, an application is being developed to automatically generate a knowledge map. With the help of automatically generated knowledge maps, students can improve the quality by using the knowledge map editing tool provided by SEA;

• combining knowledge maps: This application is intended for system administrators. After checking and accepting the knowledge maps created by users, they can combine these local knowledge maps into a global one. During this process, duplicated nodes and edges will be automatically deleted.

2.5.3. Knowledge resource management

The ultimate goal of knowledge resource management is to effectively provide large-scale knowledge resources to a large number of students. To support large-scale online learning, it is possible to create a cloud platform to support a large number of students at the same time and have implemented a distributed Hadoop file system to solve the problem of storage and access to huge resources.

2.5.4 Acquisition of knowledge by extracting knowledge units

The unit of knowledge in the developed SEA is the smallest integral object of learning, such as a definition, theorem or algorithm; it often consists of several continuous sentences. The concept of a subject area can be expanded to several units of knowledge. Thus, the concept of "computer network" corresponds to units of knowledge, including "definition of a computer network", "classification of a computer network" and "development of a computer network", since "computer network" is the basic concept (subject concept) of these units of knowledge. A learning resource (document, article) usually contains many units of knowledge, while a unit of knowledge may be present in several resources.

By providing students with the appropriate level of detail for cognition, we solve the following problems: how to effectively acquire relevant units of knowledge with rapidly growing educational resources? how to automatically detect connections between units of knowledge? The proposed method of automatic extraction of knowledge units (with associations of knowledge units) is shown in Figure 4. The general approach includes three stages:

1. Preprocessing of resources: The initial training resources can be of various types and may contain additional information. Thus, resources are preprocessed to extract texts, sentences, words and meanings of words (parts of speech) that will be useful in future stages. For Russian resources, word segmentation tools in Russian will be used to extract words and tag parts of speech (POS).

2. Extraction of concepts (with relations of concepts): extraction of knowledge units is based on knowledge of previously extracted concepts, and extraction of associations of knowledge units is based on relations of concepts. Consequently, the methods of extracting concepts and relations of concepts used in the previous work are also suitable for representing knowledge based on the KZ.

3. Extraction of knowledge units (with associations of knowledge units): Based on previously obtained information, we can extract the characteristics of knowledge units, and then use machine learning and natural language processing (NLP) methods to extract knowledge units.

Fig. 4. Structure of extraction of knowledge units

2.5.4.1. Previous work: extraction of the concept and the relationship of concepts

The extraction of concepts and relations of concepts is elementary for the extraction of units of knowledge and associations of units of knowledge. Automatic extraction of concepts from the text has been investigated by many researchers ^[28]. In SEA, learning time is not the main problem, but extraction in real-time applications should be highly efficient. Thus, we adopted a dictionary-based approach that involved two steps: first, a dictionary of concepts is created; then an index-based mapping is processed for extraction. When applied to real-time applications, concepts can be extracted by matching each noun word with an indexed dictionary. In addition, administrators can add new concepts to the dictionary from extracted candidate concepts from the new document collection.

It is necessary to ensure the relationship between concepts from the same sentence. Every two concepts in one sentence must be composed as a candidate pair. In addition, it is necessary to extract features such as order, neighboring words and contexts, and use a trained classifier to determine the relationship of concepts.

2.5.4.2. Extraction of knowledge units

A unit of knowledge is an independent basic unit for expressing holistic knowledge, as well as a basic unit for understanding knowledge. Units of knowledge often consist of one or more sentences from texts. They can be classified into different semantic types according to different standards. In SEA, it seems appropriate to implement automatic extraction of four types of knowledge units — "definition", "evolution", "causality" and "example":

? the unit of knowledge "definition" defines the meaning of a specific concept;

? the unit of knowledge "evolution" describes a process in which something gradually passes into another concept, a stage;

? the unit of knowledge "causal relationship" includes a number of actions that promote a principle or strive for a certain result;

? The unit of knowledge "example" gives concrete examples of an abstract concept or definition.

A unit of knowledge can be represented as a tuple <ID, Concepts, CoreConcept, Type, Text, DID>: where "ID" is the ordinal number of the unit of knowledge; "Concepts" is a set of concepts that appeared in the unit of knowledge; "CoreConcept" is the subject concept of the unit of knowledge, and CoreConcept?Concepts; "Type" is the semantic type of the unit of knowledge, and "Type" is {"definition", "evolution", "causal relationship", "instance"}; "Text" is the textual content of the unit of knowledge; "DID" is the ordinal number of the document in which the unit of knowledge appears. The information "ID", "Concept" and "DID" can be obtained from previous works, so the purpose of extracting knowledge units is to obtain "CoreConcept", "Type" and "Text" knowledge units from linear texts. The process of extracting knowledge units is shown in Figure 5.

Since knowledge units consist of one or more sentences, the generation of candidates for knowledge units actually occurs by grouping sentences into text fragments that focus on the same topic. Depending on whether the sentence belongs to the same text fragments as the previous sentence, the sentences can be divided into two categories — independent sentences and dependent sentences. First of all, the first sentence of each paragraph is defined as an independent sentence. For all other sentences, the following attributes are extracted:

1) the number of identical concepts between the current sentence and its previous sentence;

2) is the subject of the current sentence a pronoun;

3) the number of new concepts that have appeared in the current sentence.

In addition, a supervised learning classifier is applied to all sentences, and text fragments can be extracted according to the classification results.

Fig. 5. The process of extracting knowledge units

Different types of knowledge units have some common features, but also specific features. Common features include:

1. KZ characteristics: keywords, namely characteristic words, are the basis for the identification of knowledge units. There are functions of one word, bigrams and trigrams, "definition", "definition is" and "defined as" are examples for these functions, respectively. So, in order to apply the VSM (Vector Space Model) model to extract knowledge units, a list of keywords and the weight of each word is required. First of all, characteristic words can be roughly assembled from annotated data. In addition, positive and negative examples are used to determine the weight of each characteristic word.

2. Word and POS function (part of speech): The POS function can complement the limitations and inflexibility of the keywords function. For example, the Word-and-POS function "definition /pre" (/pre means predicative) can represent keywords such as "definition is" and "definition was". This function can also be represented using the VSM model. When applying for candidates for knowledge units, we need to pre-designate the POS of each word.

3. Position features: Different types of knowledge units are often found in different paragraph positions. So, the unit of knowledge "definition" is often found at the beginning of a paragraph, while the unit of knowledge "instance" is often found in the last part. Thus, the position function can be calculated using the formula "position =i/n", where "i" refers to the ID of the candidate per unit of knowledge, and "n" indicates the total number of candidates per unit of knowledge in the paragraph.

There are also specific features for each type of knowledge units:

1) the unit of knowledge "definition": is the first word of the candidate a concept that appeared in the title of the document;

2) unit of knowledge "example": are there continuous pause signs in the candidate's unit of knowledge (the pause sign is a punctuation mark used to highlight elements in the series);

3) the unit of knowledge "causal relationship": is its previous candidate for a unit of knowledge a unit of knowledge "definition";

4) the unit of knowledge "evolution": the number of words indicating time.

Based on the above features, a multi-class extraction model is built. To improve the extraction efficiency, you can apply the ECOC model (error-correcting output codes). According to previous experiments, the SVM classifier (Support Vector Machine) can achieve optimal results when extracting knowledge units.

Subsequently, the goal is to identify internal and hidden connections between units of knowledge. Its formal representation (<kua, kub>,kur), where <kua, kub> indicates a pair of knowledge units, and "kur" is their type of association. There are three types of associations of knowledge units, namely "pre-order", "analogy" and "illustration". "Pre-order" means that "kua" is the predecessor of "kub", "analogy" means that "kua" shows similarities in some aspects with "kub", and "illustration" means that "kua" is an explanation or example for "kub". The approach to obtaining associations from knowledge includes four stages:

1) sorting of knowledge units and assigning a global unique identifier for each knowledge unit;

2) generation of a pair of candidate associations;

3) extracting the characteristics of the concept, type and distance for each pair of candidates;

4) performing a multiclass classification to distinguish three types of associations.

To assign global identifiers, the sequence of documents is reordered in the document collection, while the order of the identifiers of knowledge units within each document is preserved. The document sorting process is based on an asymmetric distribution of the basic concepts in different documents; that is, the basic concept of the last document is often found in the text of the previous document. Consequently, the assigned sequence of identifiers often indicates the logical semantic order of knowledge units. In addition, pairs of candidate knowledge units are generated depending on the location of associations of knowledge units; that is, two knowledge units that are close in identifier are considered candidate pairs. The next step is to determine if these candidate couples really have a relationship and what type of association they have. Features including concept, type and distance were extracted from each pair of candidates, and machine learning approaches (such as SVM) were applied to perform multiclass classification, and therefore three types of associations were definitively identified among the knowledge units.

2.5.5 Presentation of the knowledge map

When designing SEA, we use a knowledge map to represent the knowledge gained about concepts, knowledge units and resources with their inherent associations at three levels: the concept level, the knowledge unit level and the resource level, as shown in Figure 6.

Fig. 6. Conceptual model of the KZ

The conceptual level is similar to ontology, which represents the concepts of the domain and the relations of concepts. The level of units of knowledge consists of units of knowledge and cognitive associations of units of knowledge, i.e. relations of "preliminary order", "analogy" or "illustration" between units of knowledge. The level of knowledge units also bridges the gap between concepts and resources, since knowledge units are related to their core concepts at the concept level and are related to their appearance at the resource level. Accordingly, knowledge maps are able to guide students with their inherent associations between knowledge within different levels of detail. As a means of providing a multi-level e-learning mode, KZ maps should be designed in such a way that they are useful both for presenting knowledge and for information management. This section discusses how to achieve these goals with the help of KZ mapping.

3. Conclusion

E-learning systems are increasingly being used to provide effective educational services in the higher education system. Many Russian universities have implemented e-learning systems.

To date, most of the literature on e-learning focuses on retention rates or the quality of content. In order to solve the problems associated with the lack of proper pedagogical support, the development of SEA for self-study should focus on human cognition, engagement, flow and motivation to promote learning-related benefits such as self-control or a sense of achievement. The SEA should be designed to cover the main components associated with the schema, including schema hierarchization, schema construction, schema automation, schema activation and/or schema interaction. Theoretically, the use of cognitive schemas can help students create, automate, refine and modify schemas, thereby creating the desired learning process applicable in all fields and contexts. Schema-based instructions with learning strategies that facilitate the development of schemas are effective not only for learning, but also for developing students' self-learning skills, such as metacognitive skills or problem-solving skills.

This article presents approaches to e-learning based on knowledge maps. Separate elements of SEA have been developed, which overcomes the traditional resource-oriented orientation of e-learning and provides its cognitive-effective orientation. SEA allows you to display knowledge in the form of a map and organize them within their inherent associations at three different levels of detail, namely at the concept level, the level of knowledge units and the level of resources. Methods of automatic generation of the knowledge map are considered.

4. Future research

High-quality learning resources and real-time educational services are essential for e-learning systems. Therefore, in the future our research will be aimed at improving the two aspects mentioned above. Firstly, in order to ensure the quality of educational resources and knowledge maps, SEA is planned to be transformed into a wiki system so that everyone can contribute to improving the quality of resources. On the other hand, it is necessary to improve the accuracy of automatic extraction. Secondly, it seems necessary to develop algorithms for extracting and displaying the knowledge map available in the cloud, which means that basic calculations can be processed distributed. In addition, methods of creating personalized knowledge maps will be studied in order to use recommendations in e-learning using a navigation approach.