Evidence Units: Ontology-Grounded Document Organization for Parser-Independent Retrieval

TL;DR

This paper introduces Evidence Units, semantically complete document chunks that improve parser-independent retrieval by grouping visual assets with contextual text, validated across multiple parsers and datasets.

Contribution

It presents an ontology-grounded schema, a global assignment algorithm, a graph-based validation layer, and cross-parser validation for constructing Evidence Units.

Findings

EU-based chunking improves retrieval LCS by +0.31.

Recall@1 increases from 0.15 to 0.51 with EU-based chunking.

Cross-parser results confirm the robustness of Evidence Units.

Abstract

Structured documents--tables paired with captions, figures with explanations, equations with the paragraphs that interpret them--are routinely fragmented when indexed for retrieval. Element-level indexing treats every parsed element as an independent chunk, scattering semantically cohesive units across separate retrieval candidates. This paper presents a parser-independent pipeline that constructs Evidence Units (EUs): semantically complete document chunks that group visual assets with their contextual text. We introduce four contributions: (1) ontology-grounded role normalization extending DoCO that maps heterogeneous parser outputs to a unified semantic schema; (2) a semantic global assignment algorithm that optimally assigns paragraphs to EUs via a full similarity matrix; (3) a graph-based decision layer in Neo4j that formalizes EU construction rules and validates completeness through two invariants; and (4) cross-parser validation showing EU spatial footprints converge across MinerU and Docling, with gains preserved under parser-induced bbox variance. Experiments on OmniDocBench v1.0 (1,340 pages; 1,551 QA pairs) show EU-based chunking improves retrieval LCS by +0.31 (0.50 to 0.81). Recall@1 increases from 0.15 to 0.51 (3.4x) and MinK decreases from 2.58 to 1.72. Cross-parser results confirm the gain (LCS +0.23 to +0.31) is preserved across parsers. Text queries show the most dramatic gain: Recall@1 rises from 0.08 to 0.47.