The document discusses metric recovery from unweighted k-nearest neighbor (k-nn) graphs, highlighting its applications in user-side recommender systems and graph neural networks (GNNs). It outlines the challenges in estimating the latent coordinates from the k-nn graphs and presents a systematic approach to address these difficulties, including the importance of edge lengths and densities. The findings suggest that GNNs can successfully recover hidden features from graph structures, even with uninformative input features.