Code-switching patterns can be an effective route to improve performance of downstream NLP applications: A case study of humour, sarcasm and hate speech detection

Here we have the code and data for the following paper: Code-switching patterns can be an effective route to improve performance of downstream NLP applications: A case study of humour, sarcasm and hate speech detection by Srijan Bansal, Vishal Garimella, Ayush Suhane, Jasabanta Patro, Animesh Mukherjee. Proceedings of ACL 2020

Our trained embedding: You can download the embedding here

How to run:

Humour

ML model:

• Baseline: run python grid_search_baseline.py from Humour/ML/
• Switching: run python grid_search_baseline_switching.py from Humour/ML/

To run HAN:

• Baseline: run python master_script_baseline_signal.py from Humour/HAN/
• Switching: run python master_script_switching_signal.py from Humour/HAN/

Hate

To run ML model:

• Baseline & Switching: run python grid_search.py from Hate/ML/

To run HAN:

• Baseline: run python grid_search_baseline.py from Hate/HAN/
• Switching: run python grid_search_switching.py from Hate/HAN/

Sarcasm

To run ML model:

• Baseline: run python classification.py from Sarcasm/ML/Baseline/
• Switching: run python classification.py from Sarcasm/ML/Switching/

To run HAN:

• Baseline: run python grid_search_baseline.py from Sarcasm/HAN/
• Switching: run python grid_search_switching.py from Sarcasm/HAN/