DICE aggregates independently encoded document chunks into a single vector to reduce evidence dilution in long-document dense retrieval, reporting gains on LongEmbed especially beyond 4k tokens.
Late chunking: Contextual chunk embeddings using long-context embedding models
13 Pith papers cite this work, alongside 4 external citations. Polarity classification is still indexing.
13
Pith papers citing it
4
external citations · Pith
abstract
Many use cases require retrieving smaller portions of text, and dense vector-based retrieval systems often perform better with shorter text segments, as the semantics are less likely to be over-compressed in the embeddings. Consequently, practitioners often split text documents into smaller chunks and encode them separately. However, chunk embeddings created in this way can lose contextual information from surrounding chunks, resulting in sub-optimal representations. In this paper, we introduce a novel method called late chunking, which leverages long context embedding models to first embed all tokens of the long text, with chunking applied after the transformer model and just before mean pooling - hence the term late in its naming. The resulting chunk embeddings capture the full contextual information, leading to superior results across various retrieval tasks. The method is generic enough to be applied to a wide range of long-context embedding models and works without additional training. To further increase the effectiveness of late chunking, we propose a dedicated fine-tuning approach for embedding models.
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
IdioLink introduces a benchmark dataset and evaluation showing that strong embedding models struggle to retrieve equivalent meanings across idiomatic and literal forms, relying on shallow cues instead.
ColChunk adaptively chunks visual document patches into contextual multi-vectors via clustering, cutting storage by over 90% while raising average nDCG@5 by 9 points.
SPIRE presents a tree-structured retrieval method using subdocuments, paths, and dual contextualization that produces higher-quality and more diverse citations than passage-based baselines on HTML QA benchmarks.
A contextual multimodal document retrieval benchmark (CMDR-Bench) and embedding model (CMDR-Embed) that jointly encodes multiple document pages and splits them into page-level representations, trained with a context-aware contrastive objective, outperforming non-contextual baselines by 13–16 nDCG@5.
Controlled ablations of 38 models find MLM superior to CLM on representation benchmarks while CLM offers better data efficiency and stability; a biphasic CLM-then-MLM schedule is optimal under fixed compute and improves when initialized from pretrained CLM models.
Multi-Prefix Embedding extracts per-chunk embeddings from a single forward pass over EOS-separated document chunks and matches via MaxSim while training only on document-level labels.
EASE-TTT creates a soft attention target from evidence chunks to guide query-side test-time adaptation, yielding higher macro-average scores than full-context, retrieval-only, and standard qTTT baselines on six LongBench QA tasks.
Empirical study claiming to be the first broad comparison of chunking methods in RAG, highlighting effectiveness, cost, and generalization limitations across scenarios.
InSemRAG combines dynamic intent-aware hybrid retrieval and semantics-preserving chunk repair in an iterative loop, yielding 2.65 F1 gain on HotPotQA and 1.5 accuracy gain on FEVER with 4.32x lower latency than Multi-Hop RAG via SLMs.
Entity-based chunk filtering reduces RAG vector index size by 25-36% with retrieval quality near baseline levels.
Cluster-based semantic chunking does not outperform fixed-size or recursive chunking for RAG on academic theses, and RAGAs faithfulness shows limited reliability in this setup.
A RAG pipeline with contextual PDF chunking, question-and-answer-aware retrieval and reranking using Qwen3 models reaches 0.96 accuracy on a Ukrainian multi-domain document QA shared task.
citing papers explorer
-
Lost in a Single Vector: Improving Long-Document Retrieval with Chunk Evidence Aggregation
DICE aggregates independently encoded document chunks into a single vector to reduce evidence dilution in long-document dense retrieval, reporting gains on LongEmbed especially beyond 4k tokens.
-
IdioLink: Retrieving Meaning Beyond Words Across Idiomatic and Literal Expressions
IdioLink introduces a benchmark dataset and evaluation showing that strong embedding models struggle to retrieve equivalent meanings across idiomatic and literal forms, relying on shallow cues instead.
-
Visual Late Chunking: An Empirical Study of Contextual Chunking for Efficient Visual Document Retrieval
ColChunk adaptively chunks visual document patches into contextual multi-vectors via clustering, cutting storage by over 90% while raising average nDCG@5 by 9 points.
-
SPIRE: Structure-Preserving Interpretable Retrieval of Evidence
SPIRE presents a tree-structured retrieval method using subdocuments, paths, and dual contextualization that produces higher-quality and more diverse citations than passage-based baselines on HTML QA benchmarks.
-
CMDR: Contextual Multimodal Document Retrieval
A contextual multimodal document retrieval benchmark (CMDR-Bench) and embedding model (CMDR-Embed) that jointly encodes multiple document pages and splits them into page-level representations, trained with a context-aware contrastive objective, outperforming non-contextual baselines by 13–16 nDCG@5.
-
Should We Still Pretrain Encoders with Masked Language Modeling?
Controlled ablations of 38 models find MLM superior to CLM on representation benchmarks while CLM offers better data efficiency and stability; a biphasic CLM-then-MLM schedule is optimal under fixed compute and improves when initialized from pretrained CLM models.
-
Improving Long-Context Retrieval with Multi-Prefix Embedding
Multi-Prefix Embedding extracts per-chunk embeddings from a single forward pass over EOS-separated document chunks and matches via MaxSim while training only on document-level labels.
-
EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering
EASE-TTT creates a soft attention target from evidence chunks to guide query-side test-time adaptation, yielding higher macro-average scores than full-context, retrieval-only, and standard qTTT baselines on six LongBench QA tasks.
-
Chunking Methods on Retrieval-Augmented Generation - Effectiveness Evaluation Against Computational Cost and Limitations
Empirical study claiming to be the first broad comparison of chunking methods in RAG, highlighting effectiveness, cost, and generalization limitations across scenarios.
-
Efficient RAG with Intent-Aware Retrieval and Semantics-Preserving Chunking
InSemRAG combines dynamic intent-aware hybrid retrieval and semantics-preserving chunk repair in an iterative loop, yielding 2.65 F1 gain on HotPotQA and 1.5 accuracy gain on FEVER with 4.32x lower latency than Multi-Hop RAG via SLMs.
-
Reducing Redundancy in Retrieval-Augmented Generation through Chunk Filtering
Entity-based chunk filtering reduces RAG vector index size by 25-36% with retrieval quality near baseline levels.
-
Evaluating Chunking Strategies for Retrieval-Augmented Generation on Academic Texts
Cluster-based semantic chunking does not outperform fixed-size or recursive chunking for RAG on academic theses, and RAGAs faithfulness shows limited reliability in this setup.
-
Qwen Goes Brrr: Off-the-Shelf RAG for Ukrainian Multi-Domain Document Understanding
A RAG pipeline with contextual PDF chunking, question-and-answer-aware retrieval and reranking using Qwen3 models reaches 0.96 accuracy on a Ukrainian multi-domain document QA shared task.