Match data2vec outputs


I’m working on my final projects and decided to make a Support Chat Bot. I want my bot to find the most suitable paragraph from a documentation for the question I ask him (I want to focus on trivial questions).

These are the things that I’ve done so far:

I’ve run facebook/data2vec-text-base on all my paragraphs and stored every output in a JSON object with an identifier. Further, I’ve run the same model on the question, and with the cosine-similarity, I tried to find the closest paragraph.

The problem is that the logic doesn’t seem to work…Do you guys have any suggestions?