A measure of relevance for given question-answer pair

Question

is there a way to measure the relevance for pairs of question-answer (for training own language model or measure real life human discussion).

For example, Q: How do you get the recipe A with high relevance score: I found it on google. A with low relevance score: Well it's a good question.

What's the interpretation of question v. answer when they are tokenized?

Thanks