Link Prediction and Hybrid Strategies for Updating a Social Graph Snapshot via a Limited API

We study the problem of updating a social network snapshot when the social system provides a limited public API to access the graph. We propose novel link prediction and hybrid strategies that beat a state-of-the-art competing graph strategy [1]. Our strategies make educated guesses for edge removals and creations based on graph “fingerprints”. We also propose an improved baseline strategy and more accurate cost analysis. Our experimental results are based on 12 large weekly snapshots, allowing us to study staleness and refresh frequency issues. Our new strategies are robust to staleness, and the runtime grows linearly as a snapshot becomes staler. Our strategies can reduce API calls by an order of magnitude with minimal error.