Improving chatbots — adding feature to recognise multiple questions in user’s query

Sumeet More
2 min readOct 11, 2020

--

This will be a micro blogpost but I hope it will be useful for you moving forward.

In 2016, I started building chatbots using Microsoft bot framework. While developing, I saw an edge case — normally we(humans) don’t ask one question at a time . Sometimes we ask multiple questions. This is where my chatbot was not able to distinguish multiple questions and was failing to answer correctly.

One example can be — hey what time is the event is there a specific dress code

we can see there are two questions but asked at same time. Punctuations are also missing which is making hard for bot to distinguish.

Today, I was playing with an awesome NLP Library called Spacy. I felt above problem might fall under sentence segmentation task of NLP(natural language processing). I gave it a try and results looked promising.

Below is one of the results.

Query from image was : which ground we are playing today did you call alex do you have extra black jersey

Result was array of different questions: which ground we are playing, did you call alex, do you have extra black jersey

I know as a human we can easily recognise different questions but it is little tricky for machines to understand it but sentence segmentation from Spacy is producing magic.

Below is the API which I developed for demo:

I hope you find this micro blogpost useful.

~ Happy Learning.

--

--

Sumeet More
Sumeet More

Written by Sumeet More

Software Engineer 2 at Microsoft | Backend Engineer and Architect| Blockchain & ML enthusiast | C#,.NET Core, Rust, Javascript and Go

No responses yet