Toward Recognizing Social Media Recommenders under Absent Recommendations: A Graph Neural Network-based Approach

Assessing social media algorithms is hindered by platform opacity and data unavailability. To address this, we introduce Social Media Recommenders Recognition under Absent Recommendations (SM-ARR) and present SM-ARR-G, a Graph Neural Network framework designed to identify active algorithms without internal access. SM-ARR-G forecasts user actions by comparing past behavior against candidate "infospheres" (simulated exposure patterns), selecting the best predictor as the explanation for observed dynamics. Initial experiments using the DBLP dataset as a proxy indicate that our approach can detect hidden recommenders. If further developed, this framework could potentially offer a valuable tool for external auditing and enhancing algorithmic explainability.