Interpretable Unsupervised Community Detection with LLM-Symbolized Structured Processes

2026-08-01

AI Summary

This work addresses the limitations of traditional community detection methods in complex graph structures and the reliance of existing deep learning approaches on labeled data and their lack of interpretability. The paper proposes LUCID, an unsupervised community discovery framework inspired by natural phase transition dynamics, which uniquely integrates large language models (LLMs) with structured symbolic reasoning to generate explicit, interpretable community partitioning rules without requiring training or labels. LUCID operates through a four-stage pipeline—local initialization, multi-factor merging, multi-granularity optimization, and global selection—combining k-ego context encoding, unsupervised node role modeling, and LLM-induced logical rules, evaluated via topological compactness and boundary clarity. Experiments demonstrate that LUCID significantly outperforms state-of-the-art unsupervised and semi-supervised baselines across multiple real-world datasets, achieving the best reported performance among unsupervised methods.

Abstract

Community detection is a fundamental task in graph analytics that aims to identify cohesive groups of entities with similar behaviors or interests. Classic objective-driven methods struggle with complex graph structures, while deep-learning approaches improve performance at the expense of interpretability and rely on labeled data and training. Large language models (LLMs), with strong reasoning capabilities and world knowledge, are promising for interpretable, label-free community detection. To leverage these strengths, we propose LUCID, an LLM-guided, interpretable, training-free, and unsupervised community detection method. Inspired by phase-transition kinetics in natural systems, where complex structures emerge through initialization, merging, refinement, and selection, LUCID is designed as a four-stage pipeline. Within this pipeline, the LLM induces formal rules that translate implicit knowledge into explicit and interpretable logical structures. Specifically, (1) the Local-View Community Initialization stage encodes local graph structures using k-ego contexts and unsupervised node roles; (2) the Multi-factor Community Merge stage uses LLM-induced rules to iteratively merge local communities; (3) the Multi-grain Community Refinement stage applies LLM-induced coarse-to-fine rules in parallel to reduce boundary noise; and (4) the Global-view Community Selection stage identifies high-quality communities based on topological compactness and boundary clarity. Extensive experiments on real-world datasets demonstrate that LUCID, as an unsupervised approach, achieves state-of-the-art performance and consistently outperforms leading unsupervised and semi-supervised baselines.

Problem

• community detection
• interpretability
• unsupervised learning
• large language models
• graph analytics

Innovation

• LLM-guided community detection
• interpretable unsupervised learning
• structured symbolic reasoning
• phase-transition-inspired pipeline
• training-free graph analytics