Project proposal for Microsoft's Code Fun Do Plus Plus 2018

The objective of our project is to build a location tagged early warning system, hosted as a web-application, for disasters by analysing the posts on social media sites like Twitter.

In the 21st century, Twitter has become the hub where breaking news appears first. The tweets by the victims and press have the first-hand information about what is happening at ground-zero. They are usually the first responders to any calamity or disaster. In this project we use these tweets to notify government of the impact of the disaster as it happens on the ground. Our project is crucial because the time taken by disaster relief agencies and government in getting to know the situation, analyzing it and identifying the most impacted area is very high. Lots of casualties can be reduced if the agencies have quick and accurate information about the most affected areas, where majority of the support infrastructure should be deployed.

For this purpose, the system continuously ingests data from social media sites like Twitter, processes it (i.e., using machine learning classification techniques), classifies it into one of the several categories (like damages, need food and resources, people stuck, casualties etc). We also use the geo tagged information from the tweets to identify the latitudes and longitudes on world map in real time. The volume of tweets originating from a particular region are analysed and if the number exceeds certain threshold then we predict a disaster, followed by notifying the relief teams and government about the location.

• It consists of five core components; collector, tagger, mapper, summarizer, notifier.
• The collector is responsible for data collection from Twitter using the Twitter streaming API.
• The collected tweets are then passed to the tagger, it is responsible for the classification of each individual tweet to one of the categories. The tagger is comprised of three modules: feature extractor, learner, and classifier.
• Then using latitudes and longitudes of the tweet, the mapper plots the tweet to the world map. Here is a demo image, where tweets are mapped to their geo-locations with different icons represent various categories of disaster.

• Important and actionable tweets are identified and a summarized report is generated by extracting knowledge from the tweets.
• When a large volume of tweets starts popping out from a small geo-location then the notifier detects that event, parses the tweets and sends the summarized report to disaster relief agencies as the warning.

We believe our solution can help government and disaster relief teams in following ways:

• Quickly developing situational awareness of disaster and analyzing where the critical infrastructure should be deployed.
• Prioritizing the resources to the most affected areas.
• By providing tactical and actionable information to disaster relief agencies. All actionable tweets (such as: 10 people stuck at location-x, etc.) will be summarized and sent to the disaster management team.