Search papers, labs, and topics across Lattice.
The paper introduces CGRA-DeBERTa, a concept-guided residual domain augmentation transformer, to improve question answering over Hadith corpora by incorporating theological priors. It customizes DeBERTa with LoRA adaptations and a residual concept-aware gating mechanism that selectively amplifies semantically critical tokens based on an Islamic Concept Dictionary. Experiments on a dataset of 42591 QA pairs from Sahih al-Bukhari and Sahih Muslim demonstrate that CGRA-DeBERTa achieves a 97.85 EM score, surpassing BERT and DeBERTa by a significant margin while maintaining computational efficiency.
Injecting domain-specific theological concepts into a DeBERTa model boosts question answering accuracy on Islamic texts by 8%, proving that targeted knowledge infusion beats brute-force scaling.
Accurate QA over classical Islamic texts remains challenging due to domain specific semantics, long context dependencies, and concept sensitive reasoning. Therefore, a new CGRA DeBERTa, a concept guided residual domain augmentation transformer framework, is proposed that enhances theological QA over Hadith corpora. The CGRA DeBERTa builds on a customized DeBERTa transformer backbone with lightweight LoRA based adaptations and a residual concept aware gating mechanism. The customized DeBERTa embedding block learns global and positional context, while Concept Guided Residual Blocks incorporate theological priors from a curated Islamic Concept Dictionary of 12 core terms. Moreover, the Concept Gating Mechanism selectively amplifies semantically critical tokens via importance weighted attention, applying differential scaling from 1.04 to 3.00. This design preserves contextual integrity, strengthens domain-specific semantic representations, and enables accurate, efficient span extraction while maintaining computational efficiency. This paper reports the results of training CGRA using a specially constructed dataset of 42591 QA pairs from the text of Sahih alBukhari and Sahih Muslim. While BERT achieved an EM score of 75.87 and DeBERTa one of 89.77, our model scored 97.85 and thus surpassed them by 8.08 on an absolute scale, all while adding approximately 8 inference overhead due to parameter efficient gating. The qualitative evaluation noted better extraction and discrimination and theological precision. This study presents Hadith QA systems that are efficient, interpretable, and accurate and that scale provide educational materials with necessary theological nuance.
