Hate Speech Detection (HaSpeeDe3)

The HaSpeeDe3 task presented at EVALITA 2023 focused at the detection of hateful content written in Italian tweets.

The following code can generate an instance of the system used in the competition.

>>> from EvoMSA.competitions import Comp2023
>>> D = # Training set
>>> tailored = 'evalita2023_HaSpeeDe3'
>>> comp2023 = Comp2023(lang='it', tailored=tailored)
>>> ins = comp2023.stack_2_bow_tailored_keywords(D)
Performance in Cross-validation

Configuration

Performance

p-value

Comp2023.stack_2_bow_tailored_keywords

0.8778

1.0000

Comp2023.stack_3_bows_tailored_keywords

0.8769

0.2720

Comp2023.stack_2_bow_keywords

0.8756

0.0500

Comp2023.stack_bow_keywords_emojis_voc_selection

0.8754

0.1380

Comp2023.bow_voc_selection

0.8751

0.1600

Comp2023.stack_bows

0.8745

0.1020

Comp2023.bow

0.8740

0.0780

Comp2023.stack_bow_keywords_emojis

0.8727

0.0080

Comp2023.stack_3_bows

0.8723

0.0260

Comp2023.bow_training_set

0.8688

0.0000