Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Metadata-Driven Retrieval-Augmented Generation for Financial Question Answering

Michail Dadopoulos, Anestis Ladas, Stratos Moschidis, Ioannis Negkakis

TL;DR

This work tackles the difficulty of grounding RAG in long, structured financial filings by introducing a metadata-driven, multi-stage RAG architecture that leverages LLM-generated document and chunk metadata to create contextual chunks. An offline indexing pipeline produces hierarchical metadata and two vector collections (Standard and Contextual), enabling pre-retrieval filtering, query rewriting, and metadata-enriched embeddings. Comprehensive experiments on FinanceBench show that a metadata-aware approach with a robust reranker and contextual chunks yields substantial gains in retrieval precision and answer faithfulness, approaching the performance of commercial systems while offering lower cost and higher interpretability. The study provides a practical blueprint for building auditable, efficient, and accurate financial document analysis systems that better handle numerical reasoning and cross-document syntheses in high-stakes settings.

Abstract

Retrieval-Augmented Generation (RAG) struggles on long, structured financial filings where relevant evidence is sparse and cross-referenced. This paper presents a systematic investigation of advanced metadata-driven Retrieval-Augmented Generation (RAG) techniques, proposing and evaluating a novel, multi-stage RAG architecture that leverages LLM-generated metadata. We introduce a sophisticated indexing pipeline to create contextually rich document chunks and benchmark a spectrum of enhancements, including pre-retrieval filtering, post-retrieval reranking, and enriched embeddings, benchmarked on the FinanceBench dataset. Our results reveal that while a powerful reranker is essential for precision, the most significant performance gains come from embedding chunk metadata directly with text ("contextual chunks"). Our proposed optimal architecture combines LLM-driven pre-retrieval optimizations with these contextual embeddings to achieve superior performance. Additionally, we present a custom metadata reranker that offers a compelling, cost-effective alternative to commercial solutions, highlighting a practical trade-off between peak performance and operational efficiency. This study provides a blueprint for building robust, metadata-aware RAG systems for financial document analysis.

Metadata-Driven Retrieval-Augmented Generation for Financial Question Answering

TL;DR

This work tackles the difficulty of grounding RAG in long, structured financial filings by introducing a metadata-driven, multi-stage RAG architecture that leverages LLM-generated document and chunk metadata to create contextual chunks. An offline indexing pipeline produces hierarchical metadata and two vector collections (Standard and Contextual), enabling pre-retrieval filtering, query rewriting, and metadata-enriched embeddings. Comprehensive experiments on FinanceBench show that a metadata-aware approach with a robust reranker and contextual chunks yields substantial gains in retrieval precision and answer faithfulness, approaching the performance of commercial systems while offering lower cost and higher interpretability. The study provides a practical blueprint for building auditable, efficient, and accurate financial document analysis systems that better handle numerical reasoning and cross-document syntheses in high-stakes settings.

Abstract

Retrieval-Augmented Generation (RAG) struggles on long, structured financial filings where relevant evidence is sparse and cross-referenced. This paper presents a systematic investigation of advanced metadata-driven Retrieval-Augmented Generation (RAG) techniques, proposing and evaluating a novel, multi-stage RAG architecture that leverages LLM-generated metadata. We introduce a sophisticated indexing pipeline to create contextually rich document chunks and benchmark a spectrum of enhancements, including pre-retrieval filtering, post-retrieval reranking, and enriched embeddings, benchmarked on the FinanceBench dataset. Our results reveal that while a powerful reranker is essential for precision, the most significant performance gains come from embedding chunk metadata directly with text ("contextual chunks"). Our proposed optimal architecture combines LLM-driven pre-retrieval optimizations with these contextual embeddings to achieve superior performance. Additionally, we present a custom metadata reranker that offers a compelling, cost-effective alternative to commercial solutions, highlighting a practical trade-off between peak performance and operational efficiency. This study provides a blueprint for building robust, metadata-aware RAG systems for financial document analysis.

Paper Structure

This paper contains 32 sections, 9 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related work
  3. Method
  4. Dataset and Evaluation Framework
  5. FinanceBench: A Testbed for Financial Reasoning
  6. RAGChecker: A Framework for Fine-Grained Evaluation
  7. Experimental Setup
  8. Large Language Models (LLMs)
  9. Embedding and Retrieval Infrastructure
  10. Metadata-Enriched Indexing Pipeline (Offline Phase)
  11. Document Parsing and Structuring
  12. Document-Level Metadata Generation
  13. Chunking and Chunk-Level Metadata Enrichment
  14. Vectorization and Storage
  15. Experimental Design - RAG Architectures Under Investigation (Online Phase)
  16. ...and 17 more sections

Figures (1)

  • Figure 1: Offline Indexing Pipeline