Keyword and question selection is an internal volume caution.
On the selection of "Keyword-Question Pairs"
The situation is an S-dimensional vector
One S dimension key is determined for each pair of keywords and questions.
Because that's the object of choice.
Just multiply this inner product by Softmax and choose
[Selection from an indefinite-length object
Concerned that the number of keywords to take depends on the question.
One S-dimensional vector per object of choice is all that is needed.
The function that generates the vector may be different for each question that takes 0, 1, or 2 keywords
https://gyazo.com/5f6df8a258cda4ba2954003d5bf2d58d
---
This page is auto-translated from /nishio/キーワードと質問の選択は内積注意. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.