I would like to use RASA Response Selectors for a QnA system where answers are multiline texts, not just single lines. Say a FAQ system allowing answers are paragraphs of documents.
Take this Q/A example, where question is one-line:
How to follow people on Twitter?
and the corresponding answer (real example from twitter FAQ) is a paragraph:
1. Find a Tweet from the account you’d like to follow.
2. Hover your mouse over their name.
3. Click the **Follow** button.
OR
1. Navigate to a Tweet from the account you’d like to follow.
2. Tap the icon located at the top of the Tweet.
3. Tap **Follow** from the selection menu.
How can I manage this composite / multi line answer with data/responses.md
Response Selector markdown format?
As far as I undersatnd (please tell me I'm wrong) now in RASA I can just have 1 line for the answer. If that's true, the only solution I found is to insert explicitly a \n
every newline in the text. So the mentioned example could be translated like that:
(venv) $ cat data/responses.md
## ask languages
* faq/ask_how_to_follow_people_on_twitter
- 1. Find a Tweet from the account you’d like to follow.\n 2. Hover your mouse over their name.\n 3. Click the **Follow** button.\n \n OR \n\n 1. Navigate to a Tweet from the account you’d like to follow.\n 2. Tap the icon located at the top of the Tweet. \n 3. Tap **Follow** from the selection menu.
## ask blablabla
* faq/ask_blablabla
- blablabla\nblablabla\nblablabla
Pretty unreadable, isn't it? It "works" but I don't like it for many reasons:
- the answers on
data/responses.md
file become unreadable - I need to write a script that convert FAQs in markdown format, substituting each newline with
\n
s. - using RASA cli commands (e.g.
rasa shell
) conversation tests, I got the unreadable single line compacted texts :/
Two questions:
Can I have FAQ answers in responses.md that are composed by multi-line texts (WTSIWYG "as" wrote in the markdown file, possibly whithout inserting
'\n'
)?What's the suggested way to manage RASA Response Selectors answers, in case of answers made by long texts (~ a page/paragraph)?
UPDATE
I opened the issue/change request: https://github.com/RasaHQ/rasa/issues/5800
Thanks for your help
giorgio