The project was an attempt to solve the JPMorgan's Chase & Co. challenge. Basically, guessing and reporting where is the emergency situation based on social media. We chose as we think the most adequate social network for this kind of information - Twitter. Currently, we have several microservices in our project. One is actively crawling new twitter posts with respect to keywords and hashtag then sends the info to the afterprocessing microservice. It decides whether we are interested in this tweet and then submits it to the storage service. Our frontend is capable of displaying the map (based on Google Maps) with pointers on it. User can see the tweet content and link to it by clicking on the marker in the map. If several markers are put together, they are clustered. Thus, from a certain scale one can see "regions" where interesting tweets appears. Also, a user can add his/her own markers with the corresponding text (help requests?..).
Accomplishments that we're proud of: it's all running! :) What's next: our ultimate goal was to develop a ANN Classifier similar to "spam" filter for tweets, but it appeared to be an extremely hard task given the 24h time frame. Things to consider here are not only the tweet content, but also the source (e.g., Police department). It is important to gather a decent size database of "interesting" tweets (we got only around 20 in couple hour and all related to hurricane harvey). Also, people usually dont use the Geolocations services in twitter, so it might be useful to parse this information from tweet when it is inside the contents.
What's happening now : we display recent tweets about Federer as a proof of work. We think the framework is decent to actually complete the whole thing if having the good data about "emergency" tweets.
Log in or sign up for Devpost to join the conversation.