SPIRE: Structure-Preserving Interpretable Retrieval of Evidence

TL;DR

This paper introduces SPIRE, a structure-preserving retrieval system for semi-structured documents like HTML, enhancing evidence retrieval by maintaining document structure and providing more interpretable, citation-ready results.

Contribution

SPIRE presents a novel, structure-aware retrieval pipeline that operates over tree-structured documents, improving evidence quality and interpretability over traditional linearized methods.

Findings

Higher-quality, diverse citations achieved with structure preservation.

Outperforms passage-based baselines on HTML question-answering benchmarks.

Maintains scalability while enhancing interpretability.

Abstract

Retrieval-augmented generation over semi-structured sources such as HTML is constrained by a mismatch between document structure and the flat, sequence-based interfaces of today's embedding and generative models. Retrieval pipelines often linearize documents into fixed-size chunks before indexing, which obscures section structure, lists, and tables, and makes it difficult to return small, citation-ready evidence without losing the surrounding context that makes it interpretable. We present a structure-aware retrieval pipeline that operates over tree-structured documents. The core idea is to represent candidates as subdocuments: precise, addressable selections that preserve structural identity while deferring the choice of surrounding context. We define a small set of document primitives--paths and path sets, subdocument extraction by pruning, and two contextualization mechanisms. Global contextualization adds the non-local scaffolding needed to make a selection intelligible (e.g., titles, headers, list and table structure). Local contextualization expands a seed selection within its structural neighborhood to obtain a compact, context-rich view under a target budget. Building on these primitives, we describe an embedding-based candidate generator that indexes sentence-seeded subdocuments and a query-time, document-aware aggregation step that amortizes shared structural context. We then introduce a contextual filtering stage that re-scores retrieved candidates using locally contextualized views. Across experiments on HTML question-answering benchmarks, we find that preserving structure while contextualizing selections yields higher-quality, more diverse citations under fixed budgets than strong passage-based baselines, while maintaining scalability.