Large-scale scalability and approximation algorithms

Establish the scalability of the centralitylocal-similarity fusion framework and develop approximation algorithms for large-scale complex networks with more than 10,000 nodes.

Background

The DomiRank implementation used in the paper relies on a matrix inversion and is evaluated only on networks with at most 453 nodes. The authors therefore identify scalability verification and approximation methods as unresolved for substantially larger networks. Addressing this problem requires reducing the computational burden of centrality calculation while preserving the predictive behavior of the fused indices.

References

Besides the generalization along the centrality dimension, several open questions of the framework deserve investigation: (1) adaptive learning mechanisms for the fusion weights \omega{C}(\cdot,\cdot)---for example, end-to-end optimization of the weighting-function parameters under supervision from network topology or node attributes, upgrading empirical values'' toautomatic tuning''; (2) the generalization of the fusion paradigm to directed, weighted, and dynamic temporal networks; (3) scalability verification and approximation-algorithm design on large-scale complex networks ($|V|>104$); (4) quantitative analysis of the normalization scheme for local similarity scores and the theoretical optimality of the weighting coefficients.

— Link prediction in complex networks via fusing node centrality and local similarity indices  (2609.09658 - Zhang et al., 9 Sep 2026) in Section 7.2, Methodological outlook