A new link prediction method to alleviate the cold-start problem based on extending common neighbor and degree centrality

The cold-start problem occurs when a new user with limited information joins the network, and it becomes challenging to predict new links in future networks. Several studies have proposed link prediction methods based on common neighbors by exploring topology information using the Triadic Closure concept. However, the common neighbor failed to predict future relations because the new user with cold-start problems was isolated and had no common neighbors. This study proposes a common neighbor enhanced by the proposed gravity of node pairs inspired by Newton's law of gravity called Degree of Gravity for Link Prediction (DGLP). The DGLP considers degree centrality, common neighbors, and distance between candidate node pairs generated by topological information in a single-layer network. The proposed DGLP was evaluated using sixteen datasets and nine benchmark methods. The evaluation results showed that DGLP could increase Area Under the Curve (AUC) values by 7.15%, and the average AUC value reached 0.819 for experiments with 10-fold cross-validation. In addition, the calculated ratio of successfully predicted and node pairs with the cold-start problem achieved 99.94%. The prediction ratio is calculated to ensure that DGLP alleviates the cold-start problem and outperforms benchmark methods.

Automated summary

The cold-start problem occurs when a new user with limited information joins the network, making it challenging to predict new links in future networks. This study proposes a link prediction method, DGLP, enhanced by the gravity of node pairs inspired by Newton's law of gravity, to address the common neighbor's failure in predicting future relations for new users with cold-start problems.