Medicines in Twitter

Published: 10 Apr 2017

This was the last project of Artificial Intelligence elective, and it was a free choosing one. Nicolás Mañan, Joel Rivas and I decided, with the help of our professor Dr. Carolina Martínez, to train a Neural Network under supervised learning and classify tweets that either demand or offer medicines in Venezuela.

For Venezuelans, it’s not a secret that the crisis of medicines (that started somewhere around 2013 and has remain until present) has led people to use Twitter as a tool for searching and offering medicines. This study was meant to facilitate the search of medications on Twitter using an Artificial Intelligence.

The neural network was trained with traing and testing sets of 6000 tuits manually classified and pre-processed according to characteristics that gave better experimental results in the classification. As a result, the best neuronal network obtained was capable to identify tweets with 84% accuracy and 92% recall for supply tics, 87% for demand and 74% for other tweets.

Source code for this project, and further results can be found in the project’s GitHub repository.