<![CDATA[Social media platforms increasingly rely on recommendation algorithms to decide what content users encounter, and these systems have become powerful, largely invisible architects of contemporary public opinion. This study conducts a systematic literature review, guided by the PRISMA 2020 protocol, to synthesize evidence on how algorithmic curation including engagement-optimized ranking, collaborative filtering, and personalization models shapes exposure diversity, polarization, and opinion dynamics on digital platforms. Twenty-five peer-reviewed and preprint sources published mainly between 2021 and 2026 were screened and analyzed, spanning computational social science, communication studies, and political science. The review identifies three convergent mechanisms through which algorithms influence public opinion: selective exposure amplification, engagement-driven emotional framing, and feedback-loop reinforcement of prior preferences. Unlike earlier filter-bubble narratives that treat algorithmic influence as uniformly polarizing, this review's novelty lies in synthesizing recent causal and quasi-experimental evidence showing that algorithmic effects are conditional, platform-specific, and often smaller than public discourse assumes, while still measurably shaping the visibility and legitimacy of divergent viewpoints. Findings further reveal a persistent gap between platform-level algorithmic audits and individual-level attitudinal outcomes. The review concludes that public opinion formation in algorithmic environments should be understood as a co-produced process between platform design and user agency, offering implications for platform governance, media literacy, and future empirical research design]]>
