Conversation NLP : Topics & Sentiment

2022, June, 09

NLPSide Project

I worked on a side project for topic and sentiment detection in conversations using Natural Language Processing techniques from Dec 2019 to Feb 2020.

Even though it did not have the outcome I was hoping, it was a good learning experience.

We had an application were internal business users could search a particular customer and view all the past conversations the customer had with support agents. Each conversation would have an account number, conversation id, and chat log. Chat log had all messages sent by the customer and support agents with timestamps.

Goal was to create an API that would take an convesation id as input and provide Customer Sentiment (Happy, Neutral, Frustrated, Angry) and Conversation Topics (Billing, Request, Complain, etc.) as output.

screenshot

I worked on the entire lifecycle from gathering data to using the predictions in another application.

  1. Gather Conversation Data
  2. Label Conversations
  3. Clean Dataset
  4. Train Models
  5. Test Models
  6. Expose Models as API
  7. Integrated Model API in Web UI