Fake Religious Text Detection and AI-Based Islamic Question Answering System (Trusted Islamic Artificial Intelligence Platform “Noor AI”)
Loading...
Files
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Unuversity Setif 1 Ferhat Abbas . Faculty of Sciences
Abstract
Islamic religious information is increasingly consumed through digital and online channels, which makes it easier for false, altered, or misattributed religious texts to circulate alongside authentic ones. At the same time, users who search for answers to Islamic questions often struggle to find responses that are both accurate and grounded in trusted sources. This thesis presents Noor AI, a Trusted Islamic Artificial Intelligence platform built to address these two related problems: religious text verification and Islamic question answering.
Noor AI follows a client-server architecture composed of a Flutter mobile application, a FastAPI backend, a set of artificial intelligence models, and a prepared Islamic knowledge base covering the Quran, Hadith, and Fatwa. For religious text verification, two approaches were implemented and compared: AraBERT, a transformer-based language model for Arabic, and a similarity-based model using TF-IDF and cosine similarity. For Islamic question answering, the system uses a Retrieval-Augmented Generation (RAG) pipeline that retrieves relevant documents from a ChromaDB vector database before generating an answer with a language model (GPT-4o, accessed through OpenRouter).
A practical evaluation using 20 manually verified religious texts (10 authentic and 10 fake) was conducted. The fake texts were collected from the Dorar website [28], while 5 authentic texts were taken from the existing hadith.csv verification dataset and 5 from external sources. AraBERT achieved 90%, while the TF-IDF similarity model achieved 85%. The RAG-based question answering pipeline and the main application features were also validated through functional testing. The results show that combining a similarity-based model, a transformer-based language model, and a retrieval-augmented question answering pipeline can support a functional Islamic AI platform. However, larger labeled datasets and continued human verification are still needed for sensitive religious content.
