Ensemble of Convolutional Neural Networks for Medicine Intake Recognition in Twitter

We present the results from our participation in the 2nd Social Media Mining for Health Applications Shared Task – Task 2. The goal of this task is to develop systems capable of recognizing mentions of medication intake in Twitter. Our best performing classification system is an ensemble of neural networks with features generated by word- and character-level convolutional neural network channels and a condensed weighted bag-of-words representation. A relatively strong performance is achieved, with an F-score of 66.3 according to the official evaluation, resulting in the 5th place in the shared task with performance close to the best systems created by other participating teams.