Algorithmic Biases in Religious Text Retrieval: The Ja'fari School as a Model

Abstract

This study investigates algorithmic biases in digital religious text retrieval systems, with a focus on the Twelver Ja'fari school of jurisprudence as an applied empirical model. The research examines the impact of natural language processing (NLP) and artificial intelligence algorithms on search and retrieval within Ja'fari primary sources—the Four Books and Bihar al-Anwar. The study identifies four core bias categories: training data bias, selection bias, representation bias, and linguistic bias. To validate these findings both theoretically and empirically, the researchers conducted an experimental study across four digital systems using five theological and jurisprudential queries. Quantitative analysis and visualizations measure retrieval accuracy. Findings confirm that Ja'fari sources represent less than 5% of major training datasets, and that general search engines return Sunni-predominant results in over 80% of queries. The paper concludes with recommendations targeting specialized digitization, tailored language models, and algorithmic fairness initiatives.