{
  "annotation_id": "feb0a4fa-a9a5-4bfe-8051-9deff1637b0d",
  "date_created": "2026-02-17T05:53:19.292000Z",
  "date_modified": "2026-02-17T05:53:19.292000Z",
  "file_hash": "eff927c40e32cf6c4c4e5cf90f58fe8c36a5fa9edf1e75e1295257cff199d7a5",
  "private": false,
  "record": {
    "abstract": "T\\\u0027ecnicas recientes de \"Contextualized Chunking\" inyectan res\\\u0027umenes para mejorar el contexto en RAG, pero introducen una \"diluci\\\u0027on vectorial\" que opaca el contenido local. Evaluando distintos ratios de inyecci\\\u0027on, demostramos una curva en \"U invertida\": una inyecci\\\u0027on moderada mejora el \"Recall\" (+18%), pero superar un umbral cr\\\u0027itico (CIR \u003e 0.4) reduce la precisi\\\u0027on en un 22% para consultas espec\\\u0027ificas. Proponemos un marco te\\\u0027orico para calcular el ratio \\\u0027optimo de inyecci\\\u0027on. --\n  Recent \"Contextualized Chunking\" techniques inject summaries to improve RAG context but introduce \"vector dilution\" drowning out local content. Evaluating various injection ratios, we demonstrate an \"inverted U\" curve: moderate injection boosts Recall (+18%), but exceeding a critical threshold (CIR \u003e 0.4) drops precision by 22% for specific queries. We propose a theoretical framework to calculate the optimal injection ratio.",
    "arxiv_id": "2601.08851",
    "authors": [
      "Alex Dantart"
    ],
    "categories": [
      "cs.CL",
      "cs.AI"
    ],
    "license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
    "title": "M\\\u0027as contexto no es mejor. Paradoja de la diluci\\\u0027on vectorial en RAG corporativos",
    "url": "https://arxiv.org/abs/2601.08851",
    "version": "v1"
  },
  "schema_id": "dorsal/arxiv",
  "source": {
    "execution_id": "9689ae36-be39-4854-8b5c-c126b3662f0e",
    "id": "arXiv Dataset",
    "type": "Model",
    "variant": "snapshot-2026-01-17",
    "version": "0.1.0"
  },
  "user_id": 1000002
}