Antidote

Drug abuse is a big problem, especially since many hospitals are not regularly stocked on drug medication. We believe that we have developed a preventative measure to help hospitals be aware of their surroundings.

This is a web based application that displays a heatmap of drug use in the local area using a machine learning algorithm that interprets twitter posts. By analyzing posts, the posts are separated based on the likelihood of developing an overdose per person. Then the location of the post is monitored as it then displayed as a heat map with nearby hospitals placed throughout the map.

The application was built using python and django. Using Google Maps API, we were able to locate the hospitals in the local area. We also used the Google Maps API to display the heat map, which uses the location of twitter posts. The twitter posts were parsed and taken using both the twitter and tweepy (for python) API's. These posts were then run through a machine learning algorthm that was trained using multiple colloquial slangs through pop culture regerences such as music lyrics. Based on the algorithm, we then separated posts based on the likelihood of the author developing an overdose. These points were then mapped out on the website.

There were many problems that we had to overcome. The biggest problem was the availability of location data of twitter posts. Because of limitations of the API, we were not able to both search for posts using keywords and location data. Therefore, we could not retrieve the location data. This caused a pretty big roadblock in our progress. Therefore, we decided to continue and develop the heatmap randomly as a proof of concept. Another problem we encountered was the source for training our machine learning algorithm. Because we wanted the algorithm to pick up local slang, we had to find a source that illustrated that. This proved to be difficult as we scoured through social media to find this. Because of limited time and API's we had to develop something based on existing data and pop culture references such as music lyrics.

We are proud of our implementation of our machine learning algorithm. It could be better trained with more data points, but it has been successful in classifying data based on it's training so far. Because of the machine learning, we are better able to classify the numerous twitter posts that we have to filter through. We also think that the data visualization is very clean.

We learned a lot about machine learning and the different implementations of numerous API's to accomplish our goal.

Antidote will continue to develop itself as features such as direct hospital integration (with popups where the hospitals are, with infodata such as phone numbers, etc).Further implementation such as filtering different overdoses and data analyses will be conducted.