Local-ring network automata and the impact of hyperbolic geometry in complex network link-prediction

Topological link-prediction can exploit the entire network topology (global methods) or only the neighbourhood (local methods) of the link to predict. Global methods are believed the best. Is this common belief well-founded? Stochastic-Block-Model (SBM) is a global method believed as one of the best link-predictors, therefore it is considered a reference for comparison. But, our results suggest that SBM, whose computational time is high, cannot in general overcome the Cannistraci-Hebb (CH) network automaton model that is a simple local-learning-rule of topological self-organization proved as the current best local-based and parameter-free deterministic rule for link-prediction. To elucidate the reasons of this unexpected result, we formally introduce the notion of local-ring network automata models and their relation with the nature of common-neighbours’ definition in complex network theory. After extensive tests, we recommend Structural-Perturbation-Method (SPM) as the new best global method baseline. However, even SPM overall does not outperform CH and in several evaluation frameworks we astonishingly found the opposite. In particular, CH was the best predictor for synthetic networks generated by the Popularity-Similarity-Optimization (PSO) model, and its performance in PSO networks with community structure was even better than using the original internode-hyperbolic-distance as link-predictor. Interestingly, when tested on non-hyperbolic synthetic networks the performance of CH significantly dropped down indicating that this rule of network self-organization could be strongly associated to the rise of hyperbolic geometry in complex networks. The superiority of global methods seems a “misleading belief” caused by a latent geometry bias of the few small networks used as benchmark in previous studies. We propose to found a latent geometry theory of link-prediction in complex networks.

💡 Research Summary

This paper challenges the widely held belief that global link‑prediction methods invariably outperform local approaches. The authors compare three representative algorithms: the Stochastic‑Block‑Model (SBM), a classic global method; the Structural‑Perturbation‑Method (SPM), a more recent global baseline; and the Cannistraci‑Hebb (CH) model, a deterministic, parameter‑free local rule derived from a newly formalized concept called “local‑ring network automata.” The local‑ring framework extends the traditional common‑neighbour count by incorporating the density of cycles (rings) formed among a node’s neighbours, thereby capturing a form of topological self‑organization that is especially pronounced in networks with latent hyperbolic geometry.

The authors first provide a rigorous mathematical definition of local‑ring automata and show how CH computes a score for each non‑existent edge as the product of neighbour‑neighbour connectivity and the connectivity of second‑order neighbours. This rule requires no parameter tuning and runs in linear time with respect to the number of edges.

Experimental evaluation proceeds along three axes. (1) Real‑world networks from biology, sociology, and technology are used to benchmark SBM, SPM, and CH under standard metrics (AUC, Precision@k, Recall@k). (2) Synthetic networks generated by the Popularity‑Similarity‑Optimization (PSO) model, which embeds nodes in a hyperbolic space, are employed to test the impact of latent geometry. Variants of PSO that add explicit community structure (PSO‑CM) are also examined. (3) Non‑hyperbolic synthetic models (Erdős‑Rényi, Barabási‑Albert, Random Geometric Graphs) serve as controls to assess the geometry dependence of each method.

Results reveal a striking pattern. In hyperbolic PSO networks, CH consistently achieves the highest scores across all metrics, even surpassing a direct hyperbolic‑distance predictor. The superiority persists when community structure is added, indicating that the local‑ring rule effectively exploits both popularity‑driven attachment and similarity‑driven proximity inherent in hyperbolic embeddings. Conversely, on non‑hyperbolic synthetic graphs, CH’s performance drops sharply, falling behind both SBM and SPM. This dichotomy suggests that CH’s success is tightly linked to the presence of latent hyperbolic geometry.

Among global methods, SPM emerges as the strongest baseline, outperforming SBM in most cases while still being computationally more demanding than CH. However, in several evaluation frameworks—particularly those involving small benchmark networks—CH matches or exceeds SPM, exposing a “latent geometry bias” in prior studies that relied on limited, non‑representative datasets. The authors argue that the perceived dominance of global methods is an artifact of this bias rather than an intrinsic advantage.

The discussion proposes a new “latent geometry theory of link prediction,” positing that the hidden curvature of a network’s underlying space determines which algorithmic paradigm is optimal. The paper concludes by recommending CH for hyperbolic‑like networks, SPM as the current best global reference, and calls for future work to (i) extend local‑ring automata to multi‑dimensional hyperbolic spaces, (ii) integrate dynamic growth models, and (iii) develop scalable implementations for real‑time large‑scale applications.