Prophetic Authorial Style Modeling for Detecting Fabricated Hadiths Using AraBERT

Abstract

Identifying the authenticity of the Hadiths attributed to the Messenger of Allah (ﷺ) is a science, known as "The Science of Hadith Terminology" (ʿIlm Muṣṭalaḥ al-Ḥadīth), This science establishes the principles and rules used to evaluate Prophetic Hadiths in terms of authenticity and acceptability by examining both the chain of narration (Isnad) and the text of the Hadith (Matn), as well as the reliability and qualifications of narrators, This research presents a deep learning–based approach for detecting fabricated (Mawdu’) Hadiths using Matn-only analysis without dependence on narrators’ chains, Our methodology focuses exclusively on the Prophetic speech (Kalām al-Nabī) (ﷺ) within the text, We fine-tune a pre-trained Arabic language model (AraBERT) on a carefully curated and balanced dataset of authentic (Sahih) and fabricated (Mawḍūʿ) Hadith texts. Experimental results show that the proposed approach succeeds in detecting fabricated hadith with an accuracy of 87%, and ROC-AUC of 0.92. This work emphasizes that the model is intended as an auxiliary analytical aid, and not a replacement for traditional scholarly verification.