Not so long ago, a couple on a train experienced an unfortunate event. The husband was having a stroke, and the wife tweeted about it. When the local officials noticed the tweet, they had the train stopped and were able to save the husband's life. This is only one occasion, what if there were many other that weren't noticed? What if we can automatically detect these incidents and alert the local police, hospitals?
What it does
This new feature of twitter that's designed only for emergencies, gives us an easy way to see emergency alerts nearby. If you are a local officer, or a loving citizen, you can see what's around and help those in need. The main function of this app is to automatically find the tweets that are life-threatening and notify the local police, hospital according to a model trained behind. Apart from that Emergency Tweets app allows you to select the current hashtags for you to stay updated with emergency signals.
The main functionality of this added feature allows users to access a section called #forEmergency where tweets filtered as general(GoFundMes and long term assistance), immediate attention and #twitterSOS. twitterSOS can be incorporated into the twitter application where users could use this to get immediate help and notify the local help teams. Using a spam blocker algorithm, tweets can instantaneously be filtered out.
How I built it
The front-end was done using React.js The full-stack is done using Django, Twitter API The Backend is Python, sklearn, nltk, Watson NLU API
The design is simple: We hit the Twitter Search API to get the tweets using queries selected after a thoughtful process. The hashtags are shown on the right, and can be displayed directly.
For flagging and color coding tweets we have a combined model.
- Sarcasm Detector finds out if a tweet is sarcastic, because we all know people like beating around the bush on Twitter (Thanks @Kaitlyn for the idea).
- Fear Detector finds out if the person is scared by using IBM Watson NLU API.
- If the tweet is not sarcastic and has fear elements, then the NLP model classifies those tweets as non-serious, police, or hospital related.
Challenges I ran into
Just in the last day one of our teammates got sick. We wish him very well. We faced challenges when it came to picking the right keywords to query the tweets that are emergency related.
Accomplishments that we're proud of
We have recreated a Twitter application with a very insightful feature. Being able to use this platform to develop a feature that can help many people around the world in need of assistance is something we're happy and proud of.
What we learnt
Teamwork. Our team consists of 5 people who didn't know each other until 3 days ago. But we all came together for a cause and incorporated our ideas together to create a working application. We also learnt the various features offered in the Twitter API that allows developers to use the platform to create innovative ideas
What's next for Emergency Tweets
We want to implement flagging the tweets for general hashtags too. This demands some time, and we have limited resources. Speed should be increased. As well making it publicly available.
- Ekrem Guzelyel (ML, Backend)
- Pranav Anand (ML, Backend)
- Soumya Vemuri (Full-Stack)
- Mugunthan Raju (Front-end, Database)
- Mert Can Bilgiç (Front-End)