Reproduce “Core-based Hierarchies for Efficient GraphRAG” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jakir Hossain, Ahmet Erdem Sarıyüce
Hossain and Sarıyüce introduce a *k*-core decomposition approach for constructing hierarchical communities in GraphRAG, addressing the inherent non-reproducibility of modularity-based clustering on sparse knowledge graphs. Their method creates more stable, efficient, and semantically aligned hierarchies, leading to more comprehensive and diverse LLM responses while significantly reducing token usage.
Key insights
- A formal proof demonstrates the "degeneracy of modularity" on sparse graphs, where an exponentially large number of near-optimal partitions causes modularity-based clustering to be non-reproducible.
- *k*-core decomposition provides a robust, deterministic, and density-aware alternative for building hierarchies that better reflect the structural and semantic cohesion of knowledge graphs.
- Finer-grained, leaf-level *k*-core communities generally lead to more comprehensive and diverse LLM-generated answers compared to coarser groupings or modularity-based methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.