PRIME: Pre-Retrieval Index Modeling Enhancement for Structured Document Retrieval-Real

Key points

• High-precision retrieval over external knowledge bases is essential for LLM agents to operate effectively.1
• However, existing Information Retrieval (IR) research primarily optimizes query-side retrievers (e.g., ColBERT, BGE) and generators (e.g., Self-RAG, GraphRAG), while reducing corpus-side representation to chunking.1
• Our evaluations show that deterministic metadata matches structure-aware chunking's retrieval quality while cutting embedding computation by 45.8 $\times$ and index vector count by 22.4 $\times$.1
• Across QASPER, BRIGHT, and NFCorpus, we further characterize three qualitative regimes that determine whether index-side optimization helps or degrades retrieval, while showing that these index-side effects are neutralized under a strong reranker.1

Sources (1)

• [1]PRIME: Pre-Retrieval Index Modeling Enhancement for Structured Document Retrieval-Real
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 09:44 AM

  We propose Pre-Retrieval Index Modeling & Enhancement (PRIME), an index-side optimization framework built on two complementary tracks: an LLM-driven Semantic Restructuring (SR) track that transforms long documents into structured, information-dense sections, and a deterministic Metadata Synthesis (MS) and Hierarchical Retrieval (HR) track that extracts semantic anchors and heading structure for co

  High-precision retrieval over external knowledge bases is essential for LLM agents to operate effectively.

Extractive summary: sentences quoted from the sources.