3 ms·

How do you do NLU? I.e. how do you understand what the customer said and what to do next.

• tumidpandora 4y ago
  tl;dr it doesn't. While I did start with a vision of presbot being a self-learning chatbot built to act as an interactive agent that would represent its owner (primarily b2c) in all sorts of situations. Based on the feedback, I realized that until that interaction is smooth, believable and closer to an actual dynamic conversation, it provides much less value. I was using a combination of https://www.nltk.org/ https://www.nltk.org/, https://spacy.io/ https://spacy.io/ and https://textblob.readthedocs.io/en/dev/ https://textblob.readthedocs.io/en/dev/ for NLP then. I pivoted to a rule-based bot focusing on lead capture via a linear conversation driven by user specified questions, more like an interactive version of a static form, with prescribed Q&A (AKA FAQs on the platform).
  • streetcat1 4y ago
    I don't want to hurt the project, but this is basically a chat bot that does not understand what is said to it?
    • tumidpandora 4y ago
      That's accurate, it doesn't. Didn't want to put out a half-baked conversation experience, so had to shift focus to a rule-based bot. I'm no ML expert by any means, just an enthusiast. That's not to say it can't be extended to support NLU, the backend is modular.
• tumidpandora 4y ago
  Forgot to add, the top level interactions (e.g., lead questions or email capture) require the user to choose an option on the Chat UI (via a button click) to respond, therefore, didn't really have a need to store the context/NLU. I've built some pre-defined responses to open text questions that rely on Levenshtein Distance and threshold (e.g., 90% similarity) for identifying appropriate response. Hope this helps. For stuff that's not defined, the bot says something like "I don't know."