How to actually build a conversational interface from ground up that allows you to customise different components? The tutorial below showcase a 10-step process:

  1. The purpose of the chatbot

  2. The ideal dialogue interactions

  3. Define the domain, intent, entity, and role hierarchy

  4. Define the dialogue state handlers

  5. Create Q&A knowledge base

  6. Create training data for different classifiers

  7. Train the classifiers

  8. Implement language parser

  9. Optimisation of Q&A

  10. Deployment

Step 1: The purpose of the chatbot

It’s extremely important to really understand why you are building a chatbot. Here’s some questions you should think about.

  1. Does the chatbot conversation resemble a real-world interaction?

  2. Does it save users time?

  3. Does it fall in the Goldilocks zone where the functionalities of the chatbot are narrow enough to have high accuracy classifiers yet broad enough to be useful for users

  4. Is it possible to get large size training data?

Step 2: The ideal dialogue interactions

Mock up the ideal dialogue interactions between the user and the chatbot. This ensures that you really think about the dialogue flows, not just on the common conversation flows but also the corner and exception cases. You should plan out the MVP decision tree of the whole conversation flow, from user initial interaction to different ways it could exit the interaction.



Data Scientist

Leave a Reply