A Theoretical Framework for Algorithmic Harm Prevention in Nigerian Social Media Content Moderation

Algorithmic content moderation systems deployed by social media platforms operating in Nigeria make consequential decisions about speech and information access that may cause disproportionate harm to Nigerian users while reflecting training biases from non-Nigerian contexts. This study developed an original theoretical framework for algorithmic harm prevention in content moderation for Nigerian social media users. A critical theory-building methodology was employed, drawing on systematic review of algorithmic harm and content moderation literature (68 publications from 2018 to 2024), empirical analysis of 4,200 Nigerian content moderation appeal cases obtained from a transparency partnership with one major platform, and structured consultations with 35 Nigerian civil society advocates, platform policy officers, and human rights researchers. Empirical analysis of appeal cases revealed systematic over-removal of political speech in Nigerian languages (Hausa removal rate 3.8 times higher than equivalent English content for political commentary), under-removal of incitement in Pidgin and local languages due to insufficient model training data, and significant false positive rates for religious speech incorrectly classified as extremism. The original Nigerian Content Moderation Harm Prevention Framework proposes four harm categories specific to the Nigerian context: language equity harm, political speech suppression, cultural knowledge gap harm, and crisis information blocking. Each category includes detection criteria, algorithmic intervention requirements, and human review escalation protocols. Expert review by 20 content moderation and AI safety specialists confirmed the framework's original contribution to platform accountability theory.

Keywords: content moderation, algorithmic harm, Nigerian social media, platform accountability, original framework