bcms-bertic-frenk-hate

Text classification model based on classla/bcms-bertic and fine-tuned on the FRANK dataset comprising of LGBT and migrant hatespeech. Only the Croatian subset of the data was used for fine-tuning and the dataset has been relabeled for binary classification (offensive or acceptable).

Fine-tuning hyperparameters

Fine-tuning was performed with simpletransformers. Beforehand a brief hyperparameter optimisation was performed and the presumed optimal hyperparameters are:


model_args = {
        "num_train_epochs": 12,
        "learning_rate": 1e-5,
        "train_batch_size": 74}

Performance

The same pipeline was run with two other models and with the same dataset. Accuracy and macro F1 score were recorded for each of the 6 fine-tuning sessions and post festum analyzed.

model	average accuracy	average macro F1
bcms-bertic-frenk-hate	0.8313	0.8219
EMBEDDIA/crosloengual-bert	0.8054	0.796
xlm-roberta-base	0.7175	0.7049

From recorded accuracies and macro F1 scores p-values were also calculated:

Comparison with crosloengual-bert:

test	accuracy p-value	macro F1 p-value
Wilcoxon	0.00781	0.00781
Mann Whithney	0.00108	0.00108
Student t-test	2.43e-10	1.27e-10

Comparison with xlm-roberta-base:

test	accuracy p-value	macro F1 p-value
Wilcoxon	0.00781	0.00781
Mann Whithney	0.00107	0.00108
Student t-test	4.83e-11	5.61e-11