dorsal/arxiv
View SchemaTopo-RAG: Topology-aware retrieval for hybrid text-table documents
| Authors | Alex Dantart, Marco Kóvacs-Navarro |
|---|---|
| Categories | |
| ArXiv ID | 2601.10215vv1 |
| URL | https://arxiv.org/abs/2601.10215 |
| License | http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Abstract
In enterprise datasets, documents are rarely pure. They are not just text, nor just numbers; they are a complex amalgam of narrative and structure. Current Retrieval-Augmented Generation (RAG) systems have attempted to address this complexity with a blunt tool: linearization. We convert rich, multidimensional tables into simple Markdown-style text strings, hoping that an embedding model will capture the geometry of a spreadsheet in a single vector. But it has already been shown that this is mathematically insufficient. This work presents Topo-RAG, a framework that challenges the assumption that "everything is text". We propose a dual architecture that respects the topology of the data: we route fluid narrative through traditional dense retrievers, while tabular structures are processed by a Cell-Aware Late Interaction mechanism, preserving their spatial relationships. Evaluated on SEC-25, a synthetic enterprise corpus that mimics real-world complexity, Topo-RAG demonstrates an 18.4% improvement in nDCG@10 on hybrid queries compared to standard linearization approaches. It's not just about searching better; it's about understanding the shape of information.
{
"annotation_id": "e8ffb174-bb46-4327-b3a3-e6a5fc61348e",
"date_created": "2026-02-17T05:53:23.605000Z",
"date_modified": "2026-02-17T05:53:23.605000Z",
"file_hash": "7bd33d3ed8fecf1ce2cb08043a63a4f64998bf94d47151455d357cb1ac89a458",
"private": false,
"record": {
"abstract": "In enterprise datasets, documents are rarely pure. They are not just text, nor just numbers; they are a complex amalgam of narrative and structure. Current Retrieval-Augmented Generation (RAG) systems have attempted to address this complexity with a blunt tool: linearization. We convert rich, multidimensional tables into simple Markdown-style text strings, hoping that an embedding model will capture the geometry of a spreadsheet in a single vector. But it has already been shown that this is mathematically insufficient.\n This work presents Topo-RAG, a framework that challenges the assumption that \"everything is text\". We propose a dual architecture that respects the topology of the data: we route fluid narrative through traditional dense retrievers, while tabular structures are processed by a Cell-Aware Late Interaction mechanism, preserving their spatial relationships. Evaluated on SEC-25, a synthetic enterprise corpus that mimics real-world complexity, Topo-RAG demonstrates an 18.4% improvement in nDCG@10 on hybrid queries compared to standard linearization approaches. It\u0027s not just about searching better; it\u0027s about understanding the shape of information.",
"arxiv_id": "2601.10215",
"authors": [
"Alex Dantart",
"Marco K\u00f3vacs-Navarro"
],
"categories": [
"cs.AI"
],
"license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
"title": "Topo-RAG: Topology-aware retrieval for hybrid text-table documents",
"url": "https://arxiv.org/abs/2601.10215",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "90257f2e-e4cb-434c-b21e-f5690c9e9818",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}