This document discusses graph sampling algorithms and their application to machine learning problems using large network data. It proposes a weighted vertex-induced snowball sampling (WVISS) algorithm that selects an initial sample of vertices based on external information and topology, then adds connected vertices and edges. WVISS is found to provide more precise estimates than competing methods. The document shows WVISS achieving higher accuracy than other algorithms in a recommendation system task using a simulated product co-purchases graph, requiring only half the sample size to reach 50% accuracy. Real-world graphs often combine scale-free and small-world properties, making them difficult to model. The efficiency of sampling strategies depends on the graph structure and properties.