A Novel Recursive Network for Irony Detection in Tweets (PyTorch)

Dataset description: SemEval-2018 task 3 - Irony detection in English tweets

1. Project report is in folder /doc

2. Install requirements first

3. Run models/task3/train for ntua implementation

4. Run /Vandad/try-bert to play with bert

5. Download ntua embeddings from their github repository by running one of their tasks and put it under "embeddings"

6. Results are in foler /out/experiments/

The follwing is our proposed architecture: