This is a consulting project I built at Insight Data Science. I was consulting with a travel marketplace startup, and they wanted me to build a chatbot for their customer service.

The data I have is over 10,000 chat transcripts with id and url associated with them, and a product database containing info such as cancellation policy or ticket retrieval info. I used LDA topic modeling and Anchored CorEx topic modeling for finding frequently asked questions in the transcripts. Overall, I was able to identify about 30% of the questions asked as FAQs, and automatically answer them or redirect them to a specialist.