Using NLP for disaster management

Kaustubh Hiware (~kaustubhhiware)




During Disasters, it's crucial to ensure the right resources reach the affected victims within time. However, an interaction with the volunteers and NGO's working in disaster relief revealed that there is a lack of co-ordination and misinformation efforts. We created a semi-automated platform which leverages social networks to identify important information in a crowd-sourced manner. Matches for resources are then suggested in an automated fashion, however we allow human supervision to dictate matches, thus making it easy to use for any and all NGO's and organizations working in Disaster mitigation. We implemented a real-time location identification system from microblogs which was 100 times faster than existing state of art tools available, like Stanford NLP.

For location identification, we compared our algorithm's performance with StanfordNLP and Spacy's NER tagged, along with a few naive approaches. We preprocessed tweets and used Hashtag segmentation, Proper Noun disambiguation, pattern based regex matching, observing distances between disaster words in the dependency parsed trees to obtain a set of candidate locations - which are then verified using gazetteers.

This project also won the college level Microsoft Codefundo++ Hackathon adn was one of top 21 teams across India, to present at Microsoft's AXLE 2019. The project is currently sponsored by Microsoft India, IIT Kharagpur and Qatar Computing Research Centre.


Basic understanding of Natural Language Processing

Content URLs:

The project first stood second runner's up in Microsoft's ASI workshop, 2017.

A paper was accepted at WWW'18, based on initial work:

In the news:

Speaker Info:

Kaustubh is a recent Computer Science graduate from IIT Kharagpur. His work is primarily focused on making the results from ML and NLP works more accessible to the general public via webapps and other applications. Kaustubh is an open source and Linux fanatic. He uses arch btw. Kaustubh is an incoming SWE at Mercari, Japan - an e-commerce platform currently operating in Japan and US.

Speaker Links:




I have previously been a mentor in open source projects such as Google Code-in, Kharagpur Winter of Code and Girlscript summer of code.


Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Intermediate
Last Updated:

Hi, Kaustubh! Thanks for submitting your proposal.
We have put together a set of best practices for proposals - please take a look.
The description of your talk doesn't give specific details of your presentation, and I insist you to update it.
Your project seems genuinely interesting. However, you need to make the proposal more descriptive from a presentation view, so that it's easy for the reviewers to follow.
A brief outline of your talk, and a 2-minute video would be appreciated.
In case of any queries, you can ping us up on Gitter.

Best regards,
(CFP coordinator)

Chaitanya Tejaswi (~CRTejaswi)

Thanks for your comment, I'll make the necessary changes.

Kaustubh Hiware (~kaustubhhiware)

Thanks for your comment, I'll make the necessary changes.

Kaustubh Hiware (~kaustubhhiware)

Login to add a new comment.