YASCA: A collective intelligence approach for community detection in complex networks

In this paper we present an original approach for community detection in complex networks. The approach belongs to the family of seed-centric algorithms. However, instead of expanding communities around selected seeds as most of existing approaches do, we explore here applying an ensemble clustering approach to different network partitions derived from ego-centered communities computed for each selected seed. Ego-centered communities are themselves computed applying a recently proposed ensemble ranking based approach that allow to efficiently combine various local modularities used to guide a greedy optimisation process. Results of first experiments on real world networks for which a ground truth decomposition into communities are known, argue for the validity of our approach.

💡 Research Summary

The paper introduces YASCA (Another Seed‑centric Community detection Algorithm), a novel community detection method that blends seed‑centric local analysis with ensemble clustering techniques. Traditional seed‑centric algorithms expand communities outward from a set of selected seed nodes, but YASCA takes a different route: for each seed it first extracts an ego‑centered community using an ensemble‑ranking greedy optimisation that combines several local modularity measures (e.g., internal density, conductance). This yields a set of bipartitions Pv = {Cv, V\Cv} for each seed v.

All these bipartitions are then merged through a cluster‑ensemble process. A consensus graph Gcons is built on the original node set; an edge (i, j) receives a weight equal to the frequency with which nodes i and j appear together in any of the local partitions. The authors retain only edges whose co‑occurrence frequency exceeds a threshold (0.5 in their experiments). The resulting weighted, possibly disconnected graph is finally processed by a standard community detection algorithm capable of handling weighted graphs – the Louvain method is used in the study.

The algorithm therefore consists of three main steps: (1) seed selection, (2) local ego‑community extraction, and (3) ensemble‑based consensus clustering. In the experimental section the authors evaluate YASCA on three well‑known benchmark networks with known ground‑truth partitions: Zachary’s Karate Club, the Dolphin social network, and the Political Books network. They compare the Normalized Mutual Information (NMI) of YASCA’s output against four state‑of‑the‑art methods: Louvain, Infomap, Walktrap, and the Girvan‑Newman edge‑betweenness modularity optimiser. Using a hybrid seed set composed of the top 25 % high‑degree nodes and the bottom 25 % low‑degree nodes, YASCA achieves NMI scores that are equal to or higher than those of the baselines, demonstrating its competitive performance even on small graphs.

The authors acknowledge several limitations and avenues for future work. The current evaluation is limited to relatively small networks; scaling to large‑scale graphs will require parallelising the local community extraction step for each seed. Moreover, the choice of seed selection strategy, the threshold for edge inclusion in the consensus graph, and the specific local modularities combined in the ensemble ranking can all influence results, suggesting the need for systematic parameter studies or adaptive mechanisms. Extending YASCA to dynamic or streaming networks, where seeds and local communities evolve over time, is also highlighted as a promising direction.

In summary, YASCA offers a fresh perspective on community detection by treating each seed’s ego‑network as an independent “expert” and then fusing these expert opinions through a weighted consensus graph. This approach leverages the strengths of local modularity optimisation while retaining the global coherence provided by ensemble clustering, making it particularly attractive for applications where local computations are cheap but a reliable global partition is required.