This repo contains the code for our paper "Is More Data Better? Re-thinking the Importance of Efficiency in Abusive Language Detection with Transformers-Based Active Learning".

The core scripts are in four groups:

• 0_data_prep contains the scripts for loading, cleaning and preparing the data for modelling.
• 1_experiment_utils contains the scripts defining key functions for the training and evaluation processes.
• 2_experiment_runners contains the base python scripts and slurm scripts for running each family of experiments. A family consists of the base classification framework -- transformers (distil-roBERTa) vs traditional/sklearn (SVM) and the modelling approach -- supervised "passive" learning over the full dataset (SL) vs active learning in iterations (AL).
• 3_submit_jobs contains the scripts to create a parameter grid of all the possible experimental runs and launch them on a server.

We use small-text==1.0.0a8 with the sklearn and transformers integration.

We ran all our experiments on the JADE 2 Cluster using the environment.yaml conda enviroment provided.

For more information, feel free to contact me: [email protected].