Arabic Dialect Identification

Used Datasets:

• NADI 2020 (The First Nuanced Arabic Dialect Identification Shared Task)
• QCRI Arabic Dialect Identification (QADI)

Results

• Results and findings of the Solving the First Nuanced Arabic Dialect Identification Shared Task (NADI), Subtask 1 (country level dialect identification):

Model	F1	ACC	Epochs	SEQ-LEN
Base Model	26 %	43.7 %	3	80
Ensemble Base Model	29.03 %	45.07 %	-	-
bert-base-arabertv02-twitter	29.015 %	46.52 %	4	80
MARBERTv2	29.504 %	48.45 %	4	80
Ensemble Model	31.477 %	50.11 %	-	-

• Results and findings on the second dataset (QADI):

Model	F1	ACC	Epochs	SEQ-LEN
Base Model	46.90 %	50.28 %	6	80
bert-base-arabertv02-twitter	54.93 %	57.26 %	6	80
MARBERTv2	56.31 %	58.83 %	6	80
Ensemble Model	58.80 %	61.11 %	-	-

---