Beyond Availability: Auditing Global South Cultural Visibility in Recommender Systems

Recommender systems increasingly mediate cultural discovery for billions of users, yet the geographic fairness of their content exposure remains understudied. This study experiments whether standard recommendation algorithms systematically reduce the visibility of Global South cultural content relative to its catalogue presence. Using the MovieLens 32M dataset enriched with production country metadata from The Movie Database (TMDB), we evaluate a Popularity baseline and Item-KNN across exposure share, coverage, and a diagnostic measure we term the Cultural Erasure Index (CEI), which quantifies the proportional visibility loss of a geographic group between catalogue availability and recommendation exposure. A missingness analysis confirms that the matched subset (14.8% of movies) accounts for 81.6% of all user interactions. Our findings show that Global South content, constituting 1.30% of the classified catalogue, receives near-zero algorithmic exposure: the Popularity model produces exact zero exposure across all 2,000 evaluated users (CEI = 1.000; 95% CI [1.000, 1.000]), while Item-KNN allocates 0.010% (CEI = 0.992; 95% CI [0.985, 0.998]), with 1,996 of 2,000 users receiving no Global South content. Structural analysis indicates that post-hoc fairness interventions are unlikely to recover this exposure. We discuss implications for cultural selfhood, arguing that persistent algorithmic invisibility of regional cultural content may disrupt the conditions for intergenerational cultural transmission.