I am trying to develop architecture for Customer-Agent smart reply system for chat bot. I am taking help of Googles paper published in 2016:
https://research.google/pubs/pub45189/
This talks about smart reply in Gmail use case scenario. And our objective is to do it for chat-bot case scenario where Agent and Customer Utterances happen.
The problem that I am facing is:
How should I structure my Input/ Feed my input (i.e, Agent-Customer Utterances) for the downstream model to get the top 3 or 5 responses? How model should understand that which is Agent and which is Customer Utterances?
Input looks like:
Agent: Welcome to XYZ. How can I assist you?
Customer: I am facing issue withe internet?
Agent: May I know your registered mobile number?
Customer: My mobile number is XXXX. My email ID is [email protected].
Customer: Please resolve my issue asap.
Customer or Agent can have more than one utterances and one Utterances can have more than 1 sentences in that session.
How the input should be fed to model so that we get desired smart reply?
Like in the above example it can be:
Agent Smart Reply 1: We are working on it Mr.XXX
Agent Smart Reply 2: Your issue is resolved. Thanks for contacting XYZ
Agent Smart Reply 3: Your issue will take some time to get resolved. Please wait for 24 -48 hrs.
How we can fed the Customer-Agent Utterances to model? Any paper with reproducible code or any new suggestion will be helpful.
Dear Pratik Ch
These articles might connect all the dots together,
1. https://www.aclweb.org/anthology/2020.nlp4convai-1.9.pdf
2. https://www.researchgate.net/publication/305998186_Smart_Reply_Automated_Response_Suggestion_for_Email
Kind Regards
Qamar Ul Islam
Dear Pratik Ch
These codes might be useful, have a look:
https://github.com/jinglescode/textual-similarity-universal-sentence-encoder
Kindly let me know if it is helpful.
Kind Regards
Qamar Ul Islam
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python