Our submission (NELEC) to the 3rd SemEval-2019 task : EmoContext. Pre-print (arXiv) available here
- Regex-based replacement for slangs/typos
- Basic data processing to increase vocabulary coverage
- Recurrent model with LSTM and GRU bidirectional cells (skip connections), attention and cyclic-LR
- emoji2vec for handling emojis in data
- The source code for NELEC is implemented in a proprietary internal Microsoft tool, and thus cannot be open-sourced. Contact either of the authors to know more.
- Implementation of NELEC in Python