AR-Sanad 280K dataset contains sanads with identified narrators. The dataset could be used to train models to identify narrators in Hadiths.
The large dataset has 279,625 artificial sanads in the training and validation sets and could be used to identify 18,298 narrators. It also has a test set of 27,056 real sanads from the 6 most famous hadith books.
The lite dataset has 39,147 artificial sanads in the training and validation sets and could be used to identify 2,222 narrators.
Narrators' information include: Full names, namings (appearance forms), Shuhra, Rank by Ibn Hajar, Birth date, etc.
We provide AraBERT models trained on the large dataset.
Frozen AraBERT: we train the classification layer only.
Tuned AraBERT: We tune the weights of both the BERT model and the classification layer.
We also provide code to test the models on our test set. In "EvaluateOnTest.ipynb".