Evaluating the Performance of Our Narrator Chain Extraction System for Hadith Analysis

Author: Md. Ashraful Haque, GTAF R&D
Date: 12/08/2025

At Greentech Apps Foundation (GTAF), we are continuously innovating to bridge classical Islamic scholarship with modern computational tools. As part of our ongoing mission to digitize and analyze Islamic texts, we recently completed a performance assessment of our narrator chain (isnād) extraction system—a key component of our AI-driven hadith analysis infrastructure.

Why Narrator Chains Matter

In hadith studies, isnāds (chains of narrators) are critical for assessing the authenticity of a narration. Accurately identifying and analyzing these chains can offer deep insights into transmission patterns, narrator credibility, and the interconnections across hadith collections.

Our system automates the extraction of narrator names from Arabic hadith texts, enabling scalable analysis across large datasets. To evaluate its performance, we selected a representative test set of 20 randomly chosen hadith narrations from Muwaṭṭaʾ Mālik.

Performance Analysis

The system achieved an average accuracy of 95%, with 15 out of 20 hadith processed with perfect extraction results. This is a major step forward in the field of computational linguistics for Islamic texts, especially given the nuanced grammar, implicit references, and descriptor-heavy phrases in hadith literature.

Detailed Evaluation Examples

Each example was manually evaluated for:

• Correct extraction of all named narrators
• Proper handling of descriptors (e.g., “مولى”, “أبوه”)
• Avoiding incorrect splits or merges in narrator names

✅ Example 1: 100% Accuracy

Original chain: مَالِك → دَاوُد بْن الْحُصَيْن → مُخْبِر → عَبْد اللَّه بْن عَبَّاس
Extracted: [“مالك”, “داود بن الحصين”, “عبد الله بن عباس”]
Comment: Correctly skipped the non-narrator “مخبر” (anonymous source).

⚠️ Example 5: 80% Accuracy

Original chain: مَالِك → أَبو النَّضْر مَوْلَى عُمَر بْن عُبَيْد اللَّه → أَبو سَلَمَة بْن عَبْد الرَّحْمَن → عَائِشَة
Extracted: [“مالك”, “أبي النضر”, “مولى عمر بن عبيد الله”, “أبي سلامة بن عبد الرحمن”, “عائشة”]
Comment: Treated “مولى عمر…” as a separate narrator instead of a descriptor.

✅ Example 8: 100% Accuracy

Original chain: يَحْيَى → مَالِك → هِشَام بْن عُرْوَة → أَبِيه (عروة) → عَائِشَة → فَاطِمَة
Extracted: [“يحيى”, “مالك”, “هشام بن عروة”, “عروة”, “عائشة”, “فاطمة بنت أبي حبيش”]
Comment: Correctly resolved “أَبِيهِ” to “عروة”.

Similar detailed analysis was performed for all 20 examples. The final calculated result:

Total Score: 19.0 / 20
Average Accuracy: 95%

Key Findings

📈 Strengths

• High accuracy on direct, standard narrator chains (15/20 at 100%)
• Correct handling of implicit references like “أَبِيهِ”
• Robust extraction of sequential narrators

⚠️ Areas for Improvement

• Descriptor Handling: Distinguish “مولى” as part of the narrator’s identity, not a new narrator
• Parallel Narrators: Detect when multiple transmitters are cited together
• Unnamed Narrators: Handle vague references like “رجل” more gracefully

Research Impact

Achieving 95% accuracy enables several powerful applications:

• Automated narrator network generation
• Transmission pattern visualization across collections
• Assistance in hadith authentication studies

Next Steps

Our team is now working on:

• Improving descriptor normalization (e.g., properly parsing “مولى فلان”)
• Handling multiple parallel narrators
• Adding contextual awareness for unnamed individuals in isnāds

These upgrades will further empower scholars to work at scale with large collections of hadith, supported by accurate, explainable computational tools.

Conclusion

This work represents a milestone in GTAF’s mission to build AI-powered tools for Islamic research. We’re proud to contribute to the growing field of computational Islamic studies and are excited to continue building systems that support both scholars and developers in exploring the Islamic intellectual tradition.

📬 Interested in using this tool or collaborating on hadith data science? Get in touch or explore our work at the GTAF Hadith App.

Tags: hadith, AI, NLP, machine learning, Islamic studies, isnad, computational linguistics, digital scholarship, Greentech Apps Foundation