mlsanityData to 14 Oct 2026

Lost in a Single Vector: Improving Long-Document Retrieval with Chunk Evidence Aggregation

arXiv 2606.18781 ↗First posted 17 Jun 2026cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Dense retrieval ranks one query vector against one document vector. On long documents, this interface can fail when a short but decisive span is weakened during document encoding before ranking. We study this failure mode as document-side early compression and introduce the Evidence Dilution Index (EDI) to measure how far a document-level representation falls below the strongest chunk-level evidence within the same gold document. Guided by this view, we propose DICE (Document Inference via Chunk Evidence), a training-free document-side strategy that splits documents into chunks, encodes them independently with a frozen model, and aggregates them back into a single vector while preserving the standard one-query-one-document interface. On LongEmbed, DICE improves retrieval across four backbones, with the largest gains on slices beyond 4k tokens: for Dream, Passkey >4k rises from 30.0 to 90.0 and Needle >4k from 23.3 to 74.0. Across 12,779 filtered samples, DICE yields lower EDI than the single-vector baseline in 92.8% of cases. These results establish document-level encoding as a practical and underexplored lever for long-document retrieval.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 22 of its 27 references that are in the map. First posted on arXiv 17 Jun 2026.

Builds on

22 of 27 references.

References · 22

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,297
  2. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
  3. 2020Dense Passage Retrieval for Open-Domain Question Answeringcited by 3,824
  4. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  5. 2024RULER: What's the Real Context Size of Your Long-Context Language Models?cited by 1,346
  6. 2022Text Embeddings by Weakly-Supervised Contrastive Pre-trainingcited by 1,253
  7. 2021Unsupervised Dense Information Retrieval with Contrastive Learningcited by 1,216
  8. 2020ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERTcited by 1,178
  9. 2024M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillationcited by 1,073
  10. 2023Towards General Text Embeddings with Multi-stage Contrastive Learningcited by 723
  11. 2023C-Pack: Packed Resources For General Chinese Embeddingscited by 578
  12. 2021ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interactioncited by 476
10 more
  1. 2024NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Modelscited by 472
  2. 2024LLM2Vec: Large Language Models Are Secretly Powerful Text Encoderscited by 404
  3. 2024Generative Representational Instruction Tuningcited by 233
  4. 2023LooGLE: Can Long-Context Language Models Understand Long Contexts?cited by 202
  5. 2024Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Modelscited by 73
  6. 2024FollowIR: Evaluating and Teaching Information Retrieval Models to Follow Instructionscited by 69
  7. 2024LongEmbed: Extending Embedding Models for Long Context Retrievalcited by 48
  8. 2025Diffusion vs. Autoregressive Language Models: A Text Embedding Perspectivecited by 23
  9. 2025Rethinking Chunk Size For Long-Document Retrieval: A Multi-Dataset Analysiscited by 17
  10. 2022SeDR: Segment Representation Learning for Long Documents Dense Retrievalcited by 6

Built on by

Not cited yet.