Political Ideology Detection in Italian Texts (PoliticIT)¶

The PoliticIT task (webpage) presented at EVALITA 2023 focused on identifying politicians’ ideology information.

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

>>> from EvoMSA.competitions import Comp2023
>>> comp2023 = Comp2023(lang='it')
>>> ins = comp2023.stack_3_bows()

Performance in Cross-validation (Gender)¶
Configuration	Performance	p-value
`Comp2023.stack_3_bows`	0.9792	1.0000
`Comp2023.stack_bows`	0.9583	0.2120
`Comp2023.stack_3_bows_tailored_keywords`	0.9583	0.2340
`Comp2023.bow_training_set`	0.9375	0.1260
`Comp2023.stack_2_bow_keywords`	0.8748	0.0200
`Comp2023.stack_2_bow_tailored_keywords`	0.8748	0.0200
`Comp2023.stack_bow_keywords_emojis`	0.8536	0.0160
`Comp2023.stack_bow_keywords_emojis_voc_selection`	0.8536	0.0160
`Comp2023.bow`	0.8307	0.0000
`Comp2023.bow_voc_selection`	0.8307	0.0000

Performance in Cross-validation (Ideology Binary)¶
Configuration	Performance	p-value
`Comp2023.bow_training_set`	1.0000	1.0000
`Comp2023.stack_3_bows`	0.9714	0.1580
`Comp2023.stack_3_bows_tailored_keywords`	0.9714	0.1580
`Comp2023.stack_bows`	0.8712	0.0200
`Comp2023.bow`	0.8487	0.0120
`Comp2023.bow_voc_selection`	0.8487	0.0120
`Comp2023.stack_2_bow_keywords`	0.8271	0.0060
`Comp2023.stack_2_bow_tailored_keywords`	0.8271	0.0060
`Comp2023.stack_bow_keywords_emojis`	0.7856	0.0040
`Comp2023.stack_bow_keywords_emojis_voc_selection`	0.7856	0.0040

Performance in Cross-validation (Ideology Multiclass)¶
Configuration	Performance	p-value
`Comp2023.stack_3_bows`	0.9834	1.0000
`Comp2023.stack_3_bows_tailored_keywords`	0.9834	1.0000
`Comp2023.bow_training_set`	0.9823	0.4100
`Comp2023.stack_bows`	0.7756	0.0020
`Comp2023.stack_bow_keywords_emojis`	0.7271	0.0000
`Comp2023.stack_2_bow_tailored_keywords`	0.7271	0.0000
`Comp2023.stack_bow_keywords_emojis_voc_selection`	0.7111	0.0000
`Comp2023.stack_2_bow_keywords`	0.7111	0.0000
`Comp2023.bow`	0.5308	0.0000
`Comp2023.bow_voc_selection`	0.5308	0.0000