Neural network. How to answer the test?

Question

Python

Hello.
I want to try to write a program in python and sklearn (or any other machine learning library) that can answer questions from the test.
What I mean:
Let's say we have several questions, each question has 3 (maybe a different number) of answer options. For each question, we must choose the option that fits best (correct).
Of course, large search engines are working on finding answers to user requests using artificial intelligence, but in this implementation we already have options, and this simplifies the task.
You can try to find the answer to the question through search engines: determine the number of matches of the text (you can even take texts from sites that pop up) with the answer options. You can ask a question with multiple answers, then see where more pages came out.
In theory, you can still use text analysis . For example, find the smallest distance between keywords, analyze the type of questions, again find keywords and determine what part of speech the word we are looking for should be.
Suppose we have decided how we will analyze the text and queries. Now we have all the data for this request, and each answer will have its own data.
How can we now understand which answer option is the most suitable? For example, you can inneural network to run all the information about the answer option, write whether it is correct or not. And only then, when using, we will look at which of the options the neural network predicted a higher chance.
In general, I will be glad if you write any information both on text and query analysis, and on working with a neural network (and text analysis can also have a neural network). If there are ready-made implementations, then it will be even better.
Thanks in advance)

Answer 1

rub until complete satisfaction
and

garrison
cap garrison cap
photo garrison
cap military photo garrison cap military
cap

I hope they broaden your horizons

Answer 2

Make up the chains of the roots of the words of the question and answer manually and give them "weights".
Then - train the NN on these data.