ALPINE: Active Link Prediction Using Network Embedding

Many real-world problems can be formalized as predicting links in a partially observed network. Examples include Facebook friendship suggestions, the prediction of protein–protein interactions, and identification hidden relationships crime Several link algorithms, notably those recently introduced using network embedding, are capable doing this by just relying on part Often, whether two nodes linked queried, albeit at substantial cost (e.g., questionnaires, wet lab experiments, or undercover work). Such additional information improve accuracy, but owing to cost, queries must made with due consideration. Thus, we argue that an active learning approach is great potential interest developed ALPINE (Active Link Prediction usIng Network Embedding), framework identifies most useful status estimating improvement accuracy gained querying it. We proposed several query strategies for use combination ALPINE, inspired optimal experimental design literature. Experimental results real data not only showed was scalable boosted far fewer queries, also shed light relative merits strategies, providing actionable guidance practitioners.