How can a task be implemented using neural networks?

Question

Python

There is a program for testing students.
Tests are generated on the basis of a questionnaire from the database.
The questionnaire determines the subject of the material, the test of its level.
Based on these tests, we determine the level / base of the student and prepare an individual plan for him, the student can change some parts of the individual plan.
Educational material and tests are constantly updated, the tests are different: one answer, two, in some tests you need to enter the answer (sentence or word).
The result of the test is an array:
[4, 'm', 'acfez', [1, 4, 5, 8], 16, ...]
one test - one array element.
An individual plan is drawn up basically like this:
if there is an element in such and such a range, and another element contains a key\words or a sheet contains such and such numbers-> then such and such material\lectures\...
Ind. is added to the plan. the design is also represented by an array
[1, 2, '
mat.an.
Question
With the base replenishment, the number of if'ov also increases.
The NA seems to be able to simplify the task, while it is not strong in this.
Is it possible to make the National Assembly select one of the prepared ind. based on the test results? training plans? Which direction to go? What do you advise?

Answer 1

First you need to define a list of items, a list of topics and map them to a list of items. For each topic, determine its complexity (on some scale). For each question, you need to determine its complexity (weight), topic and subject.
Thus, having this information, it is possible, on the basis of the correct answers to all questions of a given student, to determine which topics he knows well, which ones he does not know very well or does not know. Here it is elementary to determine the weight of those questions that he answered correctly in relation to the weight of all questions in the topic / subject. It will turn out something like "knows 42% of mathematical analysis, answered 89% correctly in the topic" how to kill yourself against a wall after 42% of matan ""
The neural network here can not be used. You can choose questions simply according to the principle "choose those topics that have less weight resolved." Here it is also desirable to set up dependencies, for example, this particular topic depends on the knowledge of such and such a topic, and when choosing questions, you need to take this into account and choose first those that do not have dependencies or their dependent topics / questions have already been resolved. If you are already doing very well, you can also attach "tags" to each question, for example, definitions, formulas, etc. and, accordingly, by the tags of the resolved ones, it will be possible to say that he will probably solve this still unresolved issue, because the tags in it are the same as those that were in the issue that he had already decided.