New Recipe for Semi-supervised Community Detection: Clique Annealing under Crystallization Kinetics

TL;DR

CLANN reframes semi-supervised community detection as a spontaneous, physics-inspired seed-growth process. It introduces a two-module architecture: a Nucleus Proposer that learns clique-based community cores using crystallization kinetics and a learning-free Transitive Annealer that grows these cores into full communities while balancing energy, cohesion, and integrity. The method relies on a 3-layer GCN encoder projected into hyperbolic space and a set of four crystallization-inspired losses to enforce coherent core formation and scalable expansion. Empirical results on 43 real-world networks show substantial improvements over state-of-the-art baselines, with strong adaptability to varied networks and favorable efficiency profiles. The work provides a principled, physics-grounded framework for scalable, accurate semi-supervised community detection.

Abstract

Semi-supervised community detection methods are widely used for identifying specific communities due to the label scarcity. Existing semi-supervised community detection methods typically involve two learning stages learning in both initial identification and subsequent adjustment, which often starts from an unreasonable community core candidate. Moreover, these methods encounter scalability issues because they depend on reinforcement learning and generative adversarial networks, leading to higher computational costs and restricting the selection of candidates. To address these limitations, we draw a parallel between crystallization kinetics and community detection to integrate the spontaneity of the annealing process into community detection. Specifically, we liken community detection to identifying a crystal subgrain (core) that expands into a complete grain (community) through a process similar to annealing. Based on this finding, we propose CLique ANNealing (CLANN), which applies kinetics concepts to community detection by integrating these principles into the optimization process to strengthen the consistency of the community core. Subsequently, a learning-free Transitive Annealer was employed to refine the first-stage candidates by merging neighboring cliques and repositioning the community core, enabling a spontaneous growth process that enhances scalability. Extensive experiments on \textbf{43} different network settings demonstrate that CLANN outperforms state-of-the-art methods across multiple real-world datasets, showcasing its exceptional efficacy and efficiency in community detection.

Figures (10)

• Figure 1: (a) Spontaneous annealing process where the subgrains grow into the crystallized grain by merging with other grains. (b) CLANN Schematic diagram. The initial clique spontaneous grows into a community by merging other cliques. (c) Analogy between community detection and annealing process.
• Figure 2: The pipeline of proposed model. Nodes with orange borders and identical labels belong to the same community. Dashed circles represent the currently predicted centers, while nodes filled in orange are those predicted to be within communities. It is worth mentioning overlapping communities are not shown for clarity, CLANN can also accommodate overlapping communities because core growth is driven solely by energy requirements, operating independently without considering previous community assignments.
• Figure 3: The connection between crystallization to community detection and the associated loss functions. We distill these principles into four core requisites: energy, consistency, interface, and integrity. S.E. and I.E., stand for stored and interface energy, respectively.
• Figure 4: Runtime analysis. Nodes with bigger sizes stand for the best performance.
• Figure 5: F1 scores of different loss weights.