<![CDATA[Social media algorithms have been widely argued to shape information exposure through personalization based on users’ interaction histories, with potential implications for filter-bubble and echo-chamber dynamics. This study examined how algorithmic variables affected the relevance of YouTube content recommendations and considered broader socio-religious implications of increasingly personalized visibility. A quantitative quasi-experimental Interrupted Time Series Design (ITSD) was implemented by creating 25 test accounts with diverse demographic profiles and interest themes (informed by APJII 2024). Each account followed the same procedure across five iterations: a keyword search was conducted, the top-10 recommended videos were recorded, and three recommendations were opened using a randomized selection rule, yielding 1,250 video observations. Data was collected via the YouTube API, manually coded for recommendation relevance, transformed into numeric variables, and cleaned using the Interquartile Range (IQR) method. A logistic regression model was estimated and validated using the Hosmer–Lemeshow test, logit-linearity checks, and Variance Inflation Factor (VIF) diagnostics. Simple Exponential Smoothing (SES) and Holt’s Linear Trend were applied to project recommendation patterns across iterations. Iteration emerged as the most influential predictor of recommendation relevance, whereas other variables showed small or non-significant effects. The model demonstrated acceptable fit and no problematic multicollinearity, and forecasting suggested increasing relevance across iterations. Overall, the results were consistent with the strengthening of viewing-history-based personalization, which may reduce informational diversity and may facilitate a shift of religious authority toward digital actors more adaptive to algorithmic visibility]]>
Copyrights © 2025
