Oral Presentation: IsnadGuard: Detecting Fabricated Chains of Narration in Hadith Transmission Networks, by Mr. Ghaleb Aldoboni (MBZUAI)

The isnad, or ordered chain of narrators through which a hadith is transmitted, has long been central to Islamic source criticism. While prior computational work on hadith authenticity has largely focused on the report text, or matn, this work studies the narrator-transmission structure itself. We introduce IsnadGuard, a matn-free framework for detecting and localizing fabricated chains in hadith transmission networks. Using Sanadset 650K, we construct a directed narrator graph and generate corruptions inspired by classical defect typologies, including narrator substitution and chain splicing. IsnadGuard scores each narrator-to-narrator transition using local graph statistics and narrator embeddings, aggregates edge scores under a noisy-OR multiple-instance formulation, and is trained jointly for chain classification, contrastive ranking, and edge localization. On 43,539 held-out chains, IsnadGuard improves AUROC from 0.727 to 0.836 and Hit@2 localization from 0.838 to 0.916 over a strong edge-frequency baseline, offering interpretable evidence about which transition makes a chain structurally suspicious.