I recently got the first place on the Workshop in Noisy User-generated Text (WNUT) 2017 shared task! My system uses a Multi-task Neural Network composed of a CNN and a Bidirectional LSTM. I used the network as a feature extractor on both the character and word level. I accounted for sequential inference using a Conditional Random Fields (CRF) classifier.
My system achieved the first place in both Entity and Surface Form categories! Now, I will be attending the EMNLP 17 conference to present my project on September 7 at the beautiful city of Copenhagen, Denmark!