Email understanding using Rasa


I am fairly enthusiastic about Rasa and all the applications it can uncover. I want to implement a particular use case of a hotel reservation bot by email. The main difference is that I expect (at least initially) to receive all the information to perform the booking in natural language inside a single email instead of a series of interactions with a bot. I still expect to perform slot filing for some pre-defined fields (hotel, num_people, date, etc.). Should I try to use NLTK or spaCy to divide the email into different sentences and analyse one by one? I would expect to only perform actions after understanding the whole email so the whole way on writing stories would be fairly different than a bot interaction. Could I use Rasa Core to write stories in such a less interactive and more information-heavy use case, or should I rely only on Rasa NLU to look at each sentence at a time and implement the remaining logic outside Rasa? Thank you for your help!

Best regards, Luís Santos

1 Like

Hi @luis, that would certainly be an interesting use for Rasa. Our focus is more on conversations where each message consists on one or two sentences, which is not really compatible with how emails work.

I think your idea of splitting the email into sentences and then taking a look at each one with Rasa NLU makes sense, but you would need to implement your custom logic to decide what the mail represents as a whole. Keep us updated if you find an approach that works!

Hey @luis, were you successful with your approach? We are considering something similar for a support bot. It is trained on phrases, but we thought it could maybe answer mails by responding to the respective sentences. I would love to hear if you achieved something :slight_smile:


We are also planning to go into this direction. Do you made any progress here? Would be very interested in learning from your experiences?