A Community-Structure Based Adaptively Optimized Link Prediction Algorithm

Since link prediction helps improve our understandings about the structure, functions, and evolution of networks, it has drawn much attention from both computer science and physical communities. Among many mainstream proposed algorithms, the common-neighbor based ones show prominent efficiency but neglect the influence of community structure. Based on the assumption that in the same communities common neighbors show closer relations with endpoints than in different communities, we hold that using community structure in link prediction can further distinguish the contributions of common neighbors, thus improving the prediction accuracy. Accordingly, we propose Community-Structure based model (CS), which controls the contributions of the common neighbors in different communities with endpoints. Experiments on twelve real-world networks show that compared with three typical common-neighbor based baselines, the CS model provides more accurate predictions.