GroNLP/mdebertav3-subjectivity-arabic

Fine-tuned mDeBERTa V3 model for subjectivity detection in newspaper sentences. This model was developed as part of the CLEF 2023 CheckThat! Lab Task 2: Subjectivity in News Articles.

The goal in this task is to detect whether a sentence is objective (OBJ) or subjective (SUBJ). A sentence is subjective if its content is based on or influenced by personal feelings, tastes, or opinions. Otherwise, the sentence is objective. (Antici et al., 2023).

The model was fine-tuned using a multilingual training and Arabic development dataset, for which the following (hyper)parameters were utilized:

Batch Size    = 16
Max Epochs    = 4
Learning Rate = 5e-5
Warmup Steps  = 500
Weight Decay  = 0

The model ranked second in the CheckThat! Lab and obtained a macro F1 of 0.78 and a SUBJ F1 of 0.64.