One Retrieval to Cover Them All: Co-occurrence-Aware Knowledge Base Reorganization for Session-Level RAG

2026-06-30 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A new method addresses the limitation of Retrieval-Augmented Generation (RAG) systems, which are typically optimized for single queries, by improving their performance on "session-level" information needs common in enterprise environments. Standard RAG systems cover only 41% of a user's session-level information need. The proposed approach reorganizes the knowledge base (KB) offline using co-occurrence-aware clustering and expands retrieval candidates via cluster neighborhoods at query time. Evaluated on WixQA, a dataset of 6,221 enterprise support articles, this method increases single-query session coverage to 58% (+17% absolute; 95% CI: [14.1, 20.4]). It also reduces retrieval calls to achieve 70% coverage by 34% and compresses the KB to 20% of its original size, with consistent results across four embedding models and six functional domains. The authors advocate for session-level coverage as the primary metric for enterprise RAG evaluation.

Key takeaway

For AI Engineers deploying RAG systems in enterprise environments, you should re-evaluate your primary performance metrics. Standard single-query recall is insufficient, as it covers only 41% of typical user session needs. Instead, prioritize session-level coverage and consider implementing co-occurrence-aware knowledge base reorganization. This method significantly boosts session coverage to 58% and reduces retrieval calls, offering a more robust and efficient solution for complex user interactions.

Key insights

Reorganizing knowledge bases with co-occurrence clustering significantly boosts session-level RAG coverage and efficiency for enterprise users.

Principles

• RAG evaluation should prioritize session-level coverage.
• Co-occurrence patterns reveal semantic relationships across KBs.
• Offline KB reorganization enhances online retrieval efficiency.

Method

Reorganize knowledge bases offline using co-occurrence-aware clustering. At query time, expand retrieval candidates by exploring cluster neighborhoods to improve session-level coverage.

In practice

• Apply co-occurrence clustering to enterprise KBs.
• Integrate cluster neighborhood expansion into RAG.
• Benchmark RAG systems using session-level metrics.

Topics

• Retrieval-Augmented Generation
• Knowledge Base Reorganization
• Session-Level Retrieval
• Co-occurrence Clustering
• Enterprise AI
• Information Retrieval Metrics

Best for: AI Architect, Research Scientist, CTO, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — Existing RAG benchmarks misrepresent real-world performance due to high document redundancy and similarity in enterprise corpora.
• The retrieval rebuild: Why hybrid retrieval intent tripled as enterprise RAG programs hit the scale wall — Enterprise RAG is shifting from simple vector search to complex hybrid retrieval for production agentic AI.
• FRAGATA: Semantic Retrieval of HPC Support Tickets via Hybrid RAG over 20 Years of Request Tracker History — Hybrid RAG systems can effectively transform decades of unstructured support ticket data into a semantically searchable knowledge base.
• RAG Is Not Dead. You’re Just Building It Wrong. — Optimizing RAG for production requires systematic engineering across all layers, prioritizing retrieval over LLM changes.
• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — Existing RAG benchmarks fail in redundant, real-world corpora, necessitating a new evaluation framework.
• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — RARE framework improves RAG evaluation by accounting for document redundancy in high-similarity real-world corpora.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.