Bipartite link prediction improvement using the effective utilization of edge betweenness centrality

Link prediction has become increasingly popular in recent years. Bipartite networks are still of interest for link prediction researchers, because many real-world networks have the bipartite property. Traditional unsupervised link prediction methods for bipartite networks only consider local properties of the network, neighborhood and degree. In this paper, we first propose a simple formula for edge betweenness centrality in bipartite networks as a global property, and then suggest a new ranking score for link prediction in bipartite networks based on the effective combination of local and global properties, BWX. We will test our proposed method with five baseline link prediction scoring functions (JI, PA, AA, RA, and CN) and five bipartite datasets. Results show the superiority of the proposed method in the case of two popular evaluation metrics, AUC and precision. Future investigation can be performed for other global properties of the networks, such as bipartite closeness centrality. Also, combination strategies of local and global network specifications may be studied more. Finally, comparing the results with state-of-the-art unsupervised algorithms can be a good future direction.