Year: 2015
Type: conference-paper
Abstract: As a participant in the W-NUT Lexical Normalization for English Tweets challenge, we use deep learning to address the constrained task.Specifically, we use a combination of two augmented feed forward neural networks, a flagger that identifies words to be normalized and a normalizer, to take in a single token at a time and output a corrected version of that token.Despite avoiding off-the-shelf tools trained on external data and being an entirely context-free model, our system still achieved an F1-score of 81.49%, comfortably surpassing the next runner up by 1.5% and trailing the second place model by only 0.26%.
DOI: 10.18653/v1/w15-4323
Authors: Samuel Leeman-Munk, James Lester, James Cox
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W2250759104