Experimental Study on Link Prediction in Unweighted and Weighted Time-Evolving Organizational Social Network

The paper focuses on link prediction in time-evolving social networks, i.e., networks whose structure changes over time. The aim of link prediction is to forecast future connections between pairs of unconnected nodes by analyzing existing connections. The main goal of the paper is to experimentally verify whether incorporating weights on links between vertices, indicating the strength of the relationship, enhances the performance of similarity-based link prediction compared to considering only the presence of the relation. A computational experiment using various similarity measures among vertices has been carried out. The study has been conducted on the organizational social network, which is based on email communication among employees in a public organization. The results confirmed that considering weights between vertices leads to better predictions, regardless of the similarity measure applied.