Density2R v1.0.0 - IEEE BigData 2025 Official Implementation

Density2R v1.0.0

This is the official implementation of:

Density2R: Efficient Document Re-Ranking via Embedding Density Over Parametric Knowledge of Large Language Models

Published at IEEE BigData 2025.

Highlights

• Efficient document re-ranking method
• Zero-shot / unsupervised reranking setting
• Designed for information retrieval and RAG pipelines
• Uses embedding density over LLM parametric knowledge
• Evaluated on standard retrieval benchmarks

Paper

DOI: 10.1109/BigData66926.2025.11401587