MODELLING SUPPORT SYSTEMS FOR SELECTING PROFESSIONS FOR APPLICANTS IN THE CONTENT OF PERSONALIZATION OF EDUCATION

Various methods are currently being used in examining the initial “START” knowledge of applicants and their placement for specialties. Studies show that applicants are placed on the decreasing principle in terms of their overall scores at universities. In this case, applicants with a high level of knowledge are placed in the prestigious specialties as medicine and law as they require high results. Though, while applying for other professions, the applicants do not perform enough results on the key disciplines for the profession, they are placed in those professions when the general results enable it. This causes them to face a number of problems while working both in education process and in the industry. To avoid this problem and to place applicants in a specialty that is more relevant to their level of knowledge, the introduction of an individual approach to the evaluation of initial level of knowledge may be more promising. This article presents a modeling of the "evaluation – placement" support system for the individual approach to assessing applicants' knowledge and positioning them in relevant specialties. The main goal of the system is to give each applicant the opportunity to choose and study the specialty that is more relevant to their knowledge and skills, as well as to analyze the results for each discipline along with the overall results. The system is implemented using fuzzy logic based artificial neural networks. The network consists of 100 neurons in the input layer, two hidden layers and one output layer. The number of neurons at the output is the same as the number of specialties taught at university.


Introduction
Universities have been performing the sacred objective to satisfy the scientific needs of society since its inception. Affected by this factor, the society is already experiencing the transition to the knowledge economy. At the time of globalization, the intellectual capital invested by universities and research institutions has become the main production factor [1]. The use of new technologies in all three areas of psychosocial activity to increase the interest in different disciplines can lead to significant results [2]. This factor is only possible in the e-university environment.
Currently there are many e-university models in the world such as on-off campus, tele-university, consortium [3]. However, despite so much development, applicants in almost all countries of the world are tested and placed based on their general knowledge, without taking into account the knowledge and skills in particular subjects required for specific specialties. This often makes applicants to be placed in the specialties where their knowledge and skills are incompatible.
In this case, the individual approach to entrants emerges during their admission to universities, which has many advantages. The personalization of education has already become one of the requirements for the transition to the virtual world [4]. Personalization in education is a key trend in e-learning. In [5], the authors use fuzzy logic in testing computer skills of students.
The introduction of intellectual systems in the learning environment contributes to the development of maximum interaction and interconnection between educational resources and learners [6]. New opportunities for facilitating academic tasks, developing smart content, personal and global education as well as enhancing educational effectiveness can be estimated as the changes in the artificial intelligence sector [7]. The main requirement for modern educational resources used in knowledge management systems includes the ability to adapt the education to specific educational tasks, competence levels, personality traits. Moreover, the neural networks can be applied as

Computer Sciences
To achieve the aim, the following 3 objectives are set and resolved: 1. Analysis of the results on the main disciplines covered by the specialties chosen by the applicants during the admission exams in the country.
2. Comparison of the results of the students of any university on the disciplines covered by the specialties they are admitted with the results of the disciplines taught at the university on the basis of those disciplines.
3. Grouping the answers of first-year students of any university on the subjects covered by the basic disciplines of the specialties they are admitted to, and entering these results as input data to the proposed model to determine the most suitable specialty offered by the model.

Materials and methods
In the process of modeling an applicant support system, the results of various surveys conducted by the State Examination Center of the Republic of Azerbaijan, the results of various surveys conducted among students and citizens, matrices and artificial neural networks are used the research materials.

Experiments 1. Exploring the results of entrance examinations
At the first stage of the survey, the survey is conducted based on the statistical data for the years 2013-2017 provided by the State Examination Center of the Republic of Azerbaijan. It is found that in some cases, although applicants perform low results on the basic disciplines of the chosen specialties, due to the high overall results, they can be placed in the chosen specialties [29].
In the second study, the results of the students admitted to Nakhchivan State University during the entrance exams are mutually compared with the results on the disciplines they studied at the university (Fig. 1).
It is revealed that the results on the subjects remain almost unchanged during the course of study, and the student can't achieve even higher scores in the university disciplines which play a key role. A survey is conducted among the students to find out how this problem is solved by the students (Fig. 2).
35 % of the surveyed students said they would like to go to additional training courses, 48 % -to study in distant form, 3 % -to study again, and 14 % -to change their specialization. The results of the analysis once again prove that modelling the intellectual decision support system is necessary for the placement of applicants in specialties. In the third stage of the study, the questions presented to the students in the entrance exam are grouped according to the disciplines and subjects they are referred to. These results are included as input data to the proposed model. As a result of the analysis of 981 results introduced to the system through the proposed model, it becomes clear that the knowledge of 231 students is more appropriate for other specialties rather than for chosen ones. In the end, the reasons for this are clarified. Table 1 presents the results of 10 students by the selection method, the specialties they have chosen and the specialties that the system considers appropriate according to their knowledge.

2. Assessment-guiding method architecture
Through the proposed method, the occupations are grouped in order to more easily identify the level of knowledge of a learner. These groups are conventionally referred to as I, II, III, IV, V groups which respectively include the specialties in the following disciplines: -group I -Mathematics, (Science) Physics, Engineering, Architecture, Agrarian studies; -group II -Economics, Management, International Relations, Regional Studies, Sociology and Geography; -group III -Humanitarian and Pedagogical; -group IV -Medicine, Chemistry, Biology, Psychology, Agriculture; -group V -Art, Music, and Sports. The main disciplines in the first specialty group are physics and mathematics, in the second specialty group -mathematics and geography, in the third specialty group -native language and history, and in the fourth specialty group -chemistry and biology.

Computer Sciences
Testing of "start" knowledge through the assessment-guiding method is mainly performed in two specialties, which can reveal the applicant's knowledge through test questions and is carried out in two stages. The structural scheme of the model is presented in Fig. 3.
At the first stage, the student is presented 10 questions in each of the 10 disciplines that cover the specialty groups. At the end of this phase, the general knowledge level of the learner is determined for compliance with the requirements. If the student's scientific potential provides the requirements, it is checked for his/her specialty. The level of general and specialty groups is determined by the academic management team of the e-university.
In the second round, the learner is presented 25 questions on each of the two core subjects that cover the specialty group. These questions mainly concern the specialties taught at the university. At the end of this phase, the most appropriate specialty(ies) is assigned to the learner. If there are more than one related specialty, the learner will be given the chance to choose.

Fig. 3.
Testing "start" knowledge of applicants and specialization system Through this system, the knowledge of the learner on the specific group or the specialty rather than the general knowledge is assessed. Thus, the questions are presented on the specialty group in this case, and in the second case, the questions are presented on the specialties with the requirements set out in the second phase. Once the results of the system are "analyzed", the system

"START" KNOWLEDGE AND SKILLS TESTS
Answering questions and presenting practical skills in the chosen specialty.
Answering questions and presenting practical skills in the chosen specialty group Answering questions in all theoretical disciplines and presentingpractical skills in practical disciplines.
SYSTEM OUTCOME ON RESULTS ANALYSİS Analysis of potential in the chosen specialty and system outcome Analysis of potential in the chosen specialty group and system outcome Analysis of potential in all disciplines and system outcome Original Research Article: full paper (2020), «EUREKA: Physics and Engineering» Number 2 Computer Sciences presents the outcome as "does not correspond" to the specialty, "__X i ___ fully corresponds to the specialty" and "__X j __" corresponds to the training course.

Characteristics of the elements of applicants and questions sets and the relationship between them
Obviously, each discipline is made up of different subjects, and the subjects are made up of topics. Proper answers to the questions that are relevant to the future specialty of learners are more important in his/her right placement in specific specialty.
Assume that the e-university requires the learner to perform 85 % results in physics for the specialties "electro-energy engineering" and "melioration and water economy building engineering". According to the current rules, the learner can apply to both specialties performing 85 % results without responding to the questions that involve the subject "electrics". However, the proposed assessment method does not allow the admission of that learner to the specialty "electro-energy engineering", but allows the admission to the specialties as "melioration and water economy building engineering" or other "electrics" subjects which is less important.
Let's assume that applicants (applying to be learners) willing to enter the e-university is presented as a set mentioned below: = is a set of questions prepared to test the applicants knowledge by specialty groups and occupations.
However, such a compilation of the set "S" is of no importance to the proposed method. In this case, it remains unknown which subject the presented question refers to.
To address this problem, along with the current number, discipline, content, answer options, correct answer or answers (for open-ended questions) and difficulty rate settings, it is also suggested to define the subjects and topics parameters for each question, which is a prerequisite for the proposed method to be sustainable and flexible (Fig. 4). Thus, the set "S" includes the sets of disciplines "F", each of which includes the sets of subjects "B", and the sets of subjects in turn include subsets of topics "M".
Obviously, the number of subjects per discipline and the number of topics per subject differ. The number of questions per topic is finite, and the number of elements in the set "S" varies depending on the subsets F, B, and M, and is generally fuzzy.
In this case, the following relationships can be written. Assume that is the set of disciplines in which the applicant's knowledge is tested, is the set of subjects, and { } , is the set of topics. since it is referred to a particular topic.

Fig. 4. Structure of the test questions table
Obviously, the number of questions on the topics differs. Each of the topics is related to a particular subject, and their numbers vary depending on the subjects. Hence, each of the sets of subjects consists of the combinations of unequal subsets M( j) y (1)- (3).
In this case, description of the question that generates set of questions is as ( ) .
a jy jy j S M B ∈ ∈ The subjects, in turn, are the unequal subsets of the sets of disciplines, and their separate combinations generate the sets of disciplines (4)- (6).
Thus, each element of the set of questions is represented as ( ) .
∈ ∈ All sets of disciplines are also assembled to generate the set of general questions "S" (7)-(9). .
Finally, each element of the set of questions is represented as (9), which meets the requirements of the proposed method for fully identifying the applicant's knowledge.
Here P is the maximum number of questions that can be included into the question bank. The question number or code come from a combination of these parameters. For example, the selected question (587) 157 S is generally the 587th element of the subject and refers to the seventh subject of the fifth topic of the first discipline. Set of questions is described in the Euler-Venn diagram in Fig. 5.
The questions are entered to the question bank according to the specified rules with the algorithm presented in Fig. 6.
Here ss is the serial number of questions on a specific topic, t is the type of question, s is the text of question, c is the correct answer to the question, M is the topic the question refers to, B is the subject the topic refers to, and F is the discipline the subject refers to. The code variable is the unique number of the question in the database generated from the combination of variables F, B, and M.  The VB relational principle is used to implement the algorithm. As seen from the block diagram description and the structural diagram of the algorithm, once the question is entered the dataset, each question enters only the subset it refers to, which prevents the unnecessary question to be presented to the applicant.

Computer Sciences
After specifying the characteristics of the elements of applicants and question sets, the requirement model A=(S) for the questions presented to the applicant can be represented by the matrix. Each line of O h here describes the candidate who is supposed to be a learner, and the column S zp describes the questions to be selected to test the candidate's knowledge in accordance with his/her claim.
The membership function, which indicates the degree of response to the questions s ha to be presented to the applicant is determined by a fuzzy set. The block diagram of the algorithm providing these conditions is presented in Fig. 7.
As seen from the algorithm, each applicant will be asked specific questions, covering each discipline, subject, and topic in accordance with his/her choice. The questions are distributed among the applicants in an ambiguous sequence.

Developing relationship function among elements of specialties' set, requirements and results matrices
The next step to ensure the sustainable operation of the proposed method is the development of a conventional set of "specialties" "I", which includes university-specific specialties.
A number of specialties can be generally described as However, the proposed method at the first stage, performs a guiding function for the specialty groups rather than the specialties. Therefore, let's compile a set of specialty groups IQ to perform the first phase of knowledge assessment with the proposed method.
The learner's knowledge on different specialties is examined in different disciplines. The relationships between the demand for each specialty group and the disciplines it includes and the real knowledge of the learner on different disciplines can be illustrated by fuzzy demand and supply models [30][31][32].
Then requirements to the level of knowledge for the specialty groups can be described as the matrix T ig (r) (11).
Here, W is the number of disciplines included into the r-th specialty group. The demand model IQ=(T ig ) for different disciplines of separate specialty groups can be compiled with two-dimensional matrix 4 .
Here, each line of İQ Tiq characterizes the individual requirements for the specialty groups. Each column Tiq 4W characterizes the disciplines required for the specialty group and their membership to the groups.
Thus, the degree of correspondence of knowledge to the disciplines of the specialty groups is determined by a membership function fuzzy set as follows (12). The results performed by each learner whose initial knowledge is tested can also be similarly described as a matrix (13).
Here, O is an element of a set of learners and represents a learner whose knowledge is tested. f g denotes the result of the learner on a specific discipline, and G is the total number of disciplines the learner responds to.
The demand model for the individual learners' knowledge in different disciplines can be designed with a two-dimensional matrix O=(N ig ). .
× → The general requirement for knowledge by specialty groups is described as a total of requirements set for various disciplines (14).
Moreover, the overall performance indicators of the learner can also be summarized in separate disciplines (15).
The correspondence of each learner's knowledge to the disciplines is assessed by the difference between the numerical value of the requirements set for the disciplines and the numerical value of the performance of the learner (16).
Guiding the learner for the specialty group is based on the condition min . U → That is, the smallest value of U is a discipline or disciplines on which the learner has high knowledge. U=0 is considered to be most appropriate. The mutual fuzzy inclusion of the requirements of specialty groups for disciplines and the learners' performance on the disciplines can be stated as follows (17).
In the second phase of the proposed method, the relationship between the set of specialties and the requirements matrix must provide other conditions. In this case, the requirements matrix is two-dimensional and follows from the appropriate combination of disciplines and subjects. Each line of the matrix characterizes the disciplines for which the result will be taken into account for the specialty, and the columns refer to the subjects by the disciplines (Fig. 8).
Let's assume that  is a set of specialties taught at university. Then, the requirements for disciplines and subjects by the specialty can be described as , 1, , 1, .
fb T t f k b n = = = Here, f specifies the disciplines covered by the specialty, and b denotes the subjects covered by the disciplines.
Thus, the demand model of the set of specialties versus the demand matrix is as I=(T) and is represented as a three-dimensional matrix: .
Here, each line of I v corresponds to a two-dimensional matrix of T fb . Here, the requirements on the specialties include the total questions to be answered on relevant disciplines, and the requirements on the disciplines include the total questions to be answered on relevant subjects, and the requirements for the subjects ; [ ]

Fig. 8. Requirements for disciplines on specialties
In the second phase, the results of the study can also be described on disciplines and subjects as a two-dimensional matrix (19).
The lines here are the general result of the learner on the discipline, and the columns are the results of answers to the questions covering the subjects related to the relevant disciplines (20)- (22).
N i , N f , N b respectively represent the examination results of the learner on subject, discipline, and specalty. N b and N f are calculated using the formulas (23) and (24). .
Here, BDC is the number of correct answers on the subject, FDC is the number of correct answers on the discipline, and FSS is the number of total questions on the discipline.
Thus, the demand model of the set of learners against the results matrix is described as a three-dimensional matrix: .
Here, each line of O h corresponds to a two-dimensional matrix of N fb .
In this case, in the set of learners of the results matrix, the membership function of is as follows (25).

Computer Sciences
The optimal situation in the second phase, as in the first phase, is that all the values of the correspondence matrix "U" are minimum, which is resulting from the difference of the requirement and result matrices.
The degree of reciprocal inclusion of the learners' results on the disciplines and the requirements for the specialties for the second phase can be summarized as follows (28).
Using these methods, the level of knowledge of the learner is achieved through an instrumental tool created through the database management system. Correctly answered questions are denoted as 1, and the incorrectly answered questions or unanswered questions are denoted as 0. As a result, these values are summarized on separate disciplines. If the result is considered quantitatively sufficient, the results are then transmitted to the neural network and processed as input data.

5. Assessment of "Start" knowledge using neural networks
Assessment of knowledge on specialty groups is implemented through a fully-connected neural network.
Obviously, neural networks are a mathematical model inspired from the biological neural networks. Artificial neural networks are computer programs that simulate the connected neural cells (neurons) in the human brain. Neural networks enable computers to recognize templates by self-training. As the human brain, the neural networks produce only approximate results; however, no other types of computer programs can do this work effectively.
3 types of neural networks are generally distinguished: 1) fully connected networks; 2) multiple neural networks; 3) poorly connected networks. The main principle in fully connected neuron networks is that each neuron in the network is connected to other neurons [12].
The input signal matrix of the presented model consists of 10 lines and 10 columns (Fig. 9).
Matrix is conventionally referred to as S since it contains the answered questions.
( )( ) where f denotes the discipline, and m -the topic. The output matrix consists of 1 column and 4 lines. The matrix is conventionally called I (specialty group).
The elements of the input and output matrices get the values 0 or 1. The network consists of input and output layers and 1 hidden layer. There are 100 neurons in the input layer, 10 -in the hidden layer, and 4 -in the output layer. The signals from each of the following 10 neurons in the input layer are concentrated in the following j-th layer of the hidden layer. As a result, the intellectual potential of learner on each discipline is determined. These results are transmitted to each layer of the output layer. According to the potential of the learner, the corresponding element of matrix I gets the value 1, while the other elements get the value 0. Fig. 9. Description of the neural network at the first phase of the assessment of the start knowledge of learner The network activation function is a sigmoid function [10]. The Sigmoid function is the most common form of activation function, and its mathematical expression is as follows: When the total input approaches -∞, the activation level approaches zero, and at very high values of the overall input, the activation level almost equals to "1".
At the second phase, namely, guiding the learner to the relevant specialty, the given neural network consists of input and output layers and two hidden layers (Fig. 10). The first hidden layer defines the subjects, in which the learner has more potential, while the second hidden layer identifies more promising specialties on the disciplines. In the output layer, the specialty(ies), in which the learner has more potential on both disciplines, are identified.

Discussions
There are 5 subjects for both disciplines in the presented network. It is accepted that 3 specialties are taught in this specialty group at the e-university. As a result of the experiments, the results obtained during the admission examinations of newly enrolled volunteer students in a group of universities were uploaded into the system as the input values, and the specialties, in which the knowledge of the learners was more relevant, were identified.
Some of these students voluntarily take part in the final exams at the end of the academic year together with the students of the specialties whose knowledge are considered appropriate by the model proposed. Their overall results in these specialties are higher than the overall results in the specialties they are currently studying, Table 2.
As the table shows, the indicators of the membership function are higher in the specialties considered appropriate by the proposed model.
The presented theoretical explanations and the survey examples give grounds to say once again that the introduction of the proposed model to the choice of specialization of applicants can lead to the development of better-quality personnel in the future.

Conclusion
At the initial phase of the study, the applicants' results during the admission examinations were studied, and the architecture and the working principle of the assessment and guiding method were presented. At the next phase, the characteristics of the elements of the sets of applicants and questions and the relationships between them were identified. The specialties, requirements to disciplines and subjects were clarified, and the requirements and the results of applicants were presented as a matrix and the relationship function between them is designed.
At the last phase, the application of a neural network for knowledge assessment was proposed, the network was established and "trained", and the sample values were entered and tested.
As the researches and practical results showed, the assessment of the applicants' knowledge through the proposed method might ensure the correct guidance of the knowledge level of the society. Taking this into consideration, it was concluded that the introduction of the model and the relevant system were appropriate.