Narrator identification by querying Sanad graph and utilizing the NarratorsKG on AR-Sanad 280K-v2 dataset.

Narrator disambiguation is a field within hadith science that studies unidentified narrators in hadith narration chains, also known as sanads. Sanads can be represented as graphs, with the nodes representing the narrators and the edges representing their relationships in the chain. The current methods for resolving the narrator disambiguation problem do not utilize the graph structure of the sanad, but by leveraging this structure, we can apply graph computational and deep learning techniques to identify narrators. This paper introduces a method that utilizes the sanad graph structure to identify all narrators in a given sanad. Our two-stage approach begins by generating a query embedding and identifying the top k narrator entities closest to the query embedding. We then use AraBERT to re-rank the top k narrators and make the final prediction. Our method achieves 94.6% accuracy on the validation set of AR-Sanad 280K dataset. Additionally, we present AR-Sanad 280K-v2, an updated dataset that represents real hadiths more accurately. [ABSTRACT FROM AUTHOR]