Multi-task-learning-for-Hate-Speech-Detection-Project

A Multi-task learning model for 5 classification tasks on Hate Speech dataset with three languages (Arabic, English, French). The model is based on sluice network (Sluice Network model (https://arxiv.org/abs/1705.08142).

Requirement:

Python 3.6

dyNET  0.0.0 and its dependencies (Follow instructions on https://dynet.readthedocs.io/en/latest/python.html)

progress

Repository structure:

(1) Python files:

	constants.py: Contains constants used across all files.
	predictors.py: Contains classes for sequence predictors and layers.
	run_sluice_net.py: Script to train, load, and evaluate SluiceNetwork.
	sluice_net.py: The main logic for the SluiceNetwork.
	utils.py: Utility methods for data processing.

(2) Model directory:
	To save and load the trained model, you need to create a directory (e.g., model/) with four sub-directories inside, whose names are exactly MTML/, MTSL/, STML/ and STSL/. Then you need to specify the directory (e.g., model/) in the --model-dir argumnet in the command line.

(3) Log directory:
	To save the log files of the training and evaluation, you need to create a directory (e.g., log/) with four sub-directories inside, whose names are exactly MTML/, MTSL/, STML/ and STSL/.Then you need to specify the directory (e.g., log/) in the --log-dir argumnet in the command line.

Example usage:

python run_sluice_net.py --dynet-autobatch 1  --dynet-gpus 3 --dynet-seed 123 \
                      --h-layers 1 \
                     --cross-stitch\
                     --num-subspaces 2 --constraint-weight 0.1 \
                     --constrain-matrices 1 2 --patience 3 \
                     --languages ar en fr --test-languages ar en fr\
                     --model-dir model/ --log-dir log/\
                     --task-names annotator_sentiment sentiment directness group target --train-dir 'directory of training set'\
                     --dev-dir 'directory of development set'\
                     --test-dir 'directory of test set'\
                     --embeds babylon --h-dim 200 --cross-stitch-init-scheme imbalanced --threshold 0.1


Note: 
	(1) The meaning of each argument can be found in run_sluice_net.py;
	(2) '--languages' argument refers to the language dataset whice will be used for training. 'test-languages' argument refers to the language dataset whice the model will be used for testing. So '--test-languages' can only be the subset of '--languages';