Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Chunk Twice, Embed Once: A Systematic Study of Segmentation and Representation Trade-offs in Chemistry-Aware Retrieval-Augmented Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Recursive 100-token chunking and retrieval-tuned embeddings are the best default recipe for chemistry retrieval, and chunking matters as much as the embedding model.

desk verdict A useful benchmark and the largest chunking–embedding sweep for chemistry retrieval, but the abstract's champion RT100-0 is not supported by the paper's own tables; the real story is that recursive chunking and retrieval-tuned embeddings win as families, not that one specific configuration consistently outperforms. read the letter →

arxiv 2506.17277 v1 pith:NI2OKBAF submitted 2025-06-13 cs.IR cs.AIphysics.chem-ph

Mahmoud Amiri , Thomas Bocklitz This is my paper
classification cs.IRcs.AIphysics.chem-ph
keywords Retrieval-AugmentedGenerationChemistryNLPChunkingstrategiesEmbeddingmodelsDenseretrievalDomain-specificlanguageFSUChemRxivQuest
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is a systematic comparison of the two most basic choices in building a chemistry question-answering system that searches documents: how to cut the documents into retrievable chunks, and which embedding model to use to represent those chunks as vectors. It claims that a recursive token chunker set to 100-token non-overlapping windows (RT100-0) is the most effective default, and that retrieval-tuned embedding models such as Nomic, BGE, and Intfloat E5 outperform domain-pretrained models such as SciBERT. The paper also introduces FSUChemRxivQuest, a new chemistry retrieval benchmark built from ChemRxiv preprint question-answer pairs, and evaluates 25 chunking configurations and 48 embedding models across three chemistry benchmarks. If the results hold, practitioners get a simple, low-cost recipe for chemistry RAG, and the field gets evidence that how you chunk can matter as much as which embedding you pick.

What carries the argument

The argument runs on a two-stage evaluation design that isolates one variable at a time. In stage one, a fixed embedding model scores 25 chunking configurations built from five chunking families; the winner is the recursive token chunker, which splits text by walking down a priority list of delimiters, paragraph breaks, newlines, sentence-ending punctuation, then spaces, so that chunk boundaries tend to respect sentence and paragraph structure, with a target window of 100 tokens and zero overlap (RT100-0). In stage two, that chunking is held fixed while 48 embedding models are compared under the Massive Text Embedding Benchmark (MTEB) retrieval protocol, so any performance difference can be attributed to representation rather than segmentation. The newly introduced FSUChemRxivQuest benchmark, with 970 queries and 1,545 query-passage relevance pairs over a 32,698-paragraph corpus, supplies the chemistry-specific testbed that the existing general-purpose benchmarks lack.

What would settle it

Take 200 queries from FSUChemRxivQuest, have chemistry researchers mark the passages that actually answer them, and recompute the rankings of the 25 chunking configurations and 48 embedding models on those human judgments; the paper's recipe survives only if RT100-0 and the retrieval-tuned embeddings still lead.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that segmentation and representation choices interact and that the best simple configuration for chemistry retrieval is recursive token-based chunking with a 100-token window and no overlap, paired with an embedding model trained for retrieval rather than for domain language modeling. The paper reports that the recursive chunker family beats fixed-token, semantic, and LLM-based chunkers on span-alignment and precision metrics, that non-overlapping recursive chunks give the best precision-recall balance with minimal index overhead, and that retrieval-optimized encoders (Nomic, BGE, Intfloat E5) consistently rank above domain-pretrained models like SciBERT, which fall outside the top-performing cluster. It further claims that the chunking configuration can account for roughly a tenfold variation in IoU, making segmentation as influential as the embedding model.

Load-bearing premise

The central claim relies on the relevance labels in the chemistry benchmarks being accurate enough that a higher retrieval score genuinely means a better answer source, since the new benchmark's labels came from an automated pipeline rather than from human chemists.

Editorial extensions

If this is right

  • Chemistry RAG builders can adopt RT100-0 (recursive token chunking, 100 tokens, no overlap) as a strong default: it combines high retrieval quality with minimal index size and no extra compute.
  • Embedding selection should favor retrieval-tuned contrastive models over domain-pretrained encoders; for noisy preprint text, precision-oriented models like Intfloat E5 v2 are the safer choice.
  • Overlap should be treated as a recall-for-precision trade: beyond roughly 20% overlap, fixed-span chunking loses precision steeply with little recall gain, and even recursive chunking shows diminishing returns.
  • Chunking parameters deserve the same reporting and optimization attention as embedding choice, since segmentation alone produced a tenfold spread in IoU across configurations.
  • The new FSUChemRxivQuest benchmark provides a chemistry-specific testbed for future dense retrieval models, with the preprint-derived corpus being harder and more discriminative than the Wikipedia-based QA tasks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension the paper leaves open: fine-tune a chemistry-pretrained encoder with a contrastive retrieval objective and compare it against the same architecture without retrieval tuning; the paper's results predict the retrieval-tuned version will win by a wide margin.
  • Because recall is already saturated near 0.94 on the QA-style benchmarks, further gains in chemistry RAG will likely come from precision and ranking components, such as rerankers, rather than new base embeddings.
  • The chunking result is stated for chemistry, but the mechanism, syntactic boundary awareness, is generic; whether RT100-0 transfers to other dense scientific corpora such as biomedicine or materials is an untested implication of the paper's own argument.
  • The benchmark's labels were produced by an automated pipeline; a small human-annotated subset would tell whether the reported model ordering is robust to label noise or partly an artifact of fuzzy matching.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper reports a large-scale empirical study of chunking strategies and embedding models for retrieval in chemistry-focused RAG systems. The authors evaluate 25 chunking configurations across five method families using a fixed OpenAI embedding model, and then evaluate 48 embedding models on three chemistry retrieval benchmarks, including the newly introduced FSUChemRxivQuest dataset. They claim that recursive token-based chunking with a 100-token window and no overlap (RT100-0) consistently outperforms other chunking approaches, that retrieval-tuned embeddings such as Nomic, BGE, and Intfloat E5 substantially outperform domain-pretrained models like SciBERT, and that chunking choice has an impact comparable to or greater than embedding choice. The paper releases datasets, evaluation code, and empirical benchmarks.

Significance. If the findings hold, the study would provide useful practical guidance and a reusable benchmark for chemistry-focused retrieval systems. The breadth of the evaluation (25 chunking configurations, 48 embedding models, three benchmarks) is a genuine strength, as is the use of the MTEB protocol, fixed model versions for reproducibility, and the release of code and data. The central empirical finding that retrieval-tuned contrastive embeddings outperform domain-pretrained models such as SciBERT is both plausible and well supported by the reported tables. However, the headline claim about RT100-0 being the consistently best chunking configuration is not supported by the paper's own numbers, and the absence of statistical significance testing undermines several of the paper's recommendations. The paper is a valuable contribution to the empirical literature, but its claims need to be substantially recalibrated to match the evidence.

major comments (4)
  1. [Sections 2.3-2.4, Tables 7, 9, 10] The central claim that RT100-0 'consistently outperforms' other chunking configurations is not supported by the reported numbers. Table 7 shows that RT64-16 achieves the highest mean IoU (0.123±0.093) and PrecisionΩ (0.432±0.147) among all 15 configurations, while RT100-0 (Table 9) has IoU 0.090±0.067 and PrecisionΩ 0.365±0.149. Table 10 shows that RT100-60 has a slightly higher F2 score (0.5639) than RT100-0 (0.5637), and RT100-20 ties RT100-0 on IoU (0.090). These differences are small relative to the reported standard deviations, yet no significance test is reported anywhere, despite the Introduction's promise of 'rigorous statistical significance testing' (Section 1). The Abstract and Section 5.4 should be revised to present RT100-0 as an efficient, competitive default rather than a statistically validated best, unless paired significance tests over queries (e.g., bootstrap or permutation tests) are provided.
  2. [Sections 4.3-4.4, Figure 1] The chunking evaluation appears to be conducted on a single corpus, ChemRxivQuest, which is the authors' own dataset derived from ChemRxiv preprints, yet Figure 1 and parts of Section 2.3 refer to performance 'across corpora' and 'across multiple chemistry corpora.' If the chunking conclusions rest on a single benchmark, the generality of the recommendation is correspondingly limited. In addition, the relevance judgments in ChemRxivQuest (and hence in FSUChemRxivQuest) are generated by a pipeline using GPT-4o question generation and fuzzy string matching (Section 4.1; reference [18]); no human validation or inter-annotator agreement is reported. The paper should either provide evidence of label quality or explicitly qualify the chunking results as specific to this automatically constructed benchmark.
  3. [Section 4.5.2] The metric Precision@Ω is used as a primary criterion for chunking comparisons (Tables 7 and 9), but the value of the threshold Ω is never specified in the experimental sections. The definition in Section 4.5.2 only offers a generic example ('matching spans within a 10-token window'), leaving the actual experimental setting as an unspecified free parameter. Without a concrete value for Ω, the reported PrecisionΩ numbers are not reproducible and cannot be meaningfully compared across configurations.
  4. [Section 5.3] The claim that chunking has an impact 'comparable to, or greater than' that of the embedding model is not established by the experimental design. Chunking is varied in Stage 1 with a fixed embedding model, and embeddings are varied in Stage 2 with a fixed chunker; the two factors are never jointly varied in a factorial design. Moreover, the supporting comparison contrasts a tenfold range in IoU (a span-matching metric) with a 35-percentage-point spread in Main Score (a ranking metric), which are not commensurable quantities. A joint analysis on a common metric, or an explicit statistical comparison, would be needed to support the claim.
minor comments (6)
  1. [Abstract and Section 5.4] The shorthand 'R100-0' in the Abstract and in Section 5.4 does not match the 'RT100-0' notation used in the experimental sections; please standardize the notation throughout.
  2. [Tables 4, 11, 12] The dataset name is consistently misspelled as 'FSUChemRxivQest' (missing the 'r') in Table 4, Table 11, and Table 12; this should be corrected to 'FSUChemRxivQuest'.
  3. [Figure 2 caption] The caption refers to 'R64-16' where the configuration is elsewhere called 'RT64-16'; please fix the label.
  4. [Table 3] The entry 'Sent_paraMPNetB_v2' appears twice in the model list; please deduplicate.
  5. [Section 2.1 and Section 2.3] The paper states that the grid search comprised 25 unique configurations, but Table 7 lists only 15 configurations; the relationship between the two sets (and the status of the 10 overlap-specific configurations in Table 2) should be clarified in the text.
  6. [Section 5.5] The limitations section appropriately notes that reranking, multimodal retrieval, and generative accuracy are not evaluated; given this, the title phrase 'Chemistry-Aware Retrieval-Augmented Generation' overstates the scope, since no generation quality is measured. Consider adjusting the title or framing.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: core claims are anchored on external benchmarks; the only self-citational element is the provenance of the new FSUChemRxivQuest benchmark, which is minor and not load-bearing.

  1. other [Section 2.2 (FSUChemRxivQuest) and Section 4.1 (ChemRxivQuest provenance)]
    "We introduce FSUChemRxivQuest, a specialized benchmark within the Massive Text Embedding Benchmark framework for evaluating dense retrieval systems in chemistry. Derived from ChemRxivQuest[18], this dataset repurposes question-answer pairs from ChemRxiv, a chemistry preprint repository, into retrieval-oriented queries."

    The new benchmark is not built from external human relevance judgments but repurposes the authors' own prior ChemRxivQuest [18], whose QA pairs were generated by GPT-4o and verified by fuzzy string matching. Thus one of the three evaluation pillars rests on a self-cited, self-generated ground truth. This is a minor provenance self-citation rather than a load-bearing circular reduction: the paper's main comparative claims (recursive chunking advantage; retrieval-tuned embeddings over SciBERT) are independently anchored on ChemHotpotQARetrieval and ChemNQRetrieval. It does not fit patterns 1-6 exactly, hence 'other'.

full rationale

The derivation chain is largely self-contained. Stage 1 grid-searches chunking configurations with a fixed OpenAI embedding evaluated on the ChemRxivQuest-derived corpus; Stage 2 fixes the winning configuration (RT100-0) and benchmarks 48 embeddings. The two headline conclusions - recursive token chunking is a strong default, and retrieval-tuned embeddings (Nomic, BGE, Intfloat E5) beat domain-pretrained models like SciBERT - are supported on ChemHotpotQARetrieval and ChemNQRetrieval, which are external MTEB-formatted tasks not constructed by the authors. No fitted parameter is renamed as a prediction, no equation reduces to its own input, no uniqueness theorem is imported from the authors, and no known result is repackaged under new coordinates. The only self-referential element is FSUChemRxivQuest deriving its relevance pairs from the authors' earlier ChemRxivQuest dataset; because that benchmark is one of three and the central empirical patterns replicate on the external tasks, this is a minor self-citation rather than a circularity that forces the results. The paper's internal inconsistencies around RT100-0 versus RT64-16 or RT100-60, and the absence of reported significance tests, are correctness and statistical-support concerns, not circularity. Overall score 2 reflects the minor self-citation without load-bearing circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim depends on two hand-chosen settings (chunk size 100, overlap 0, and the PrecisionOmega 10-token window), on the reliability of the in-house benchmark's automatically generated qrels, and on the assumption that retrieval metrics proxy RAG quality. No physical or formal entities are invented.

free parameters (2)
  • RT100-0 chunking configuration = chunk_size=100, overlap=0
    Selected as optimal from a grid search on the same three benchmarks later used for embedding evaluation; the choice is a hand-tuned hyperparameter, and the embedding rankings depend on it.
  • PrecisionOmega window threshold = 10 tokens
    Precision@Omega is defined with a distance threshold (e.g., 10-token window in Section 4.5.2); the threshold is chosen by the authors and affects the reported PrecisionOmega margins.
assumptions (4)
  • domain assumption The automatically generated QA pairs in ChemRxivQuest, validated by fuzzy string matching, provide accurate ground-truth relevance judgments for FSUChemRxivQuest.
    Invoked in Section 4.1 and 4.8 when constructing the new benchmark; if the qrels are noisy, model rankings on FSUChemRxivQuest could change.
  • domain assumption Retrieval performance on ChemHotpotQARetrieval, ChemNQRetrieval, and FSUChemRxivQuest is a valid proxy for chemistry RAG system quality.
    Section 5.5 acknowledges generative accuracy is not evaluated; the practical recommendations assume retrieval metrics transfer to end-to-end RAG.
  • domain assumption Mean pooling of the final hidden layer is an appropriate pooling strategy for all 48 models unless documented otherwise.
    Section 2.2 states this pooling rule; several models (e.g., Nomic, BGE v1.5) have documented task-specific pooling that may differ, potentially affecting scores.
  • standard math Exact nearest-neighbor search with cosine similarity is the retrieval backend, and the MTEB protocol applies.
    Section 4.6; this is standard practice for dense retrieval and not contested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Chunk Twice, Embed Once: A Systematic Study of Segmentation and Representation Trade-offs in Chemistry-Aware Retrieval-Augmented Generation." pith.science (2026). https://pith.science/paper/NI2OKBAF

@misc{pith2026250617277,
  author       = {Pith},
  title        = {Pith review of: Chunk Twice, Embed Once: A Systematic Study of Segmentation and Representation Trade-offs in Chemistry-Aware Retrieval-Augmented Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NI2OKBAF}},
  note         = {Machine review of arXiv:2506.17277}
}
read the original abstract

Retrieval-Augmented Generation (RAG) systems are increasingly vital for navigating the ever-expanding body of scientific literature, particularly in high-stakes domains such as chemistry. Despite the promise of RAG, foundational design choices -- such as how documents are segmented and represented -- remain underexplored in domain-specific contexts. This study presents the first large-scale, systematic evaluation of chunking strategies and embedding models tailored to chemistry-focused RAG systems. We investigate 25 chunking configurations across five method families and evaluate 48 embedding models on three chemistry-specific benchmarks, including the newly introduced QuestChemRetrieval dataset. Our results reveal that recursive token-based chunking (specifically R100-0) consistently outperforms other approaches, offering strong performance with minimal resource overhead. We also find that retrieval-optimized embeddings -- such as Nomic and Intfloat E5 variants -- substantially outperform domain-specialized models like SciBERT. By releasing our datasets, evaluation framework, and empirical benchmarks, we provide actionable guidelines for building effective and efficient chemistry-aware RAG systems.

Figures

Figures reproduced from arXiv: 2506.17277 by the authors.

Figure 1
Figure 1. Corpus-level performance distributions for chunking strategies across four retrieval [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Comparison of chunking strategy performance: (a) Precision-recall trade-offs with [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Corpus-level retrieval metric distributions across chunking methods. Subplots show [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparative analysis of retrieval performance. (a) Trade-offs between recall and pre [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Comparison of the top-15 embedding models across three chemistry retrieval [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Scatter plots of Precision@10 vs Recall@10 for the top 10 models per dataset, ranked [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Distribution of model performance across three chemistry-focused retrieval [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Principal Component Analysis (PCA) visualization of embedding model performance [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. RAG Strategies for Natural Language-Based SQL Query and REST API Call Generation

    cs.SE 2026-02 conditional novelty 5.0 of 10

    In a 631-case enterprise benchmark, iterative retrieval (CoRAG) achieved 10.29% exact match vs 7.45% for standard RAG on combined SQL/API generation with hybrid documentation, while no-RAG gives 0% exact match.

Reference graph

Works this paper leans on

123 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [18]

    Chemrxivquest: A curated chemistry question- answer database extracted from chemrxiv preprints.arXiv preprint arXiv:2505.05232, 2025

    Mahmoud Amiri and Thomas Bocklitz. Chemrxivquest: A curated chemistry question- answer database extracted from chemrxiv preprints.arXiv preprint arXiv:2505.05232, 2025

    arXiv 2025

  2. [1]

    Repro- ducibility in chemistry research.Heliyon, 10(14), 2024

    Rosaria Ciriminna, Giuseppe Angellotti, Giovanna Li Petri, and Mario Pagliaro. Repro- ducibility in chemistry research.Heliyon, 10(14), 2024

    2024

  3. [2]

    Retrieval-augmented generation for knowledge-intensive nlp tasks.ArXiv, abs/2005.11401, 2020

    Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Kuttler, Mike Lewis, Wen tau Yih, Tim Rocktaschel, Sebastian Riedel, and Douwe Kiela. Retrieval-augmented generation for knowledge-intensive nlp tasks.ArXiv, abs/2005.11401, 2020. URLhttps://api.semanticscholar. org/CorpusID:218869575

    arXiv 2005

  4. [3]

    Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997, 2:1, 2023

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yixin Dai, Jiawei Sun, Haofen Wang, and Haofen Wang. Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997, 2:1, 2023

    arXiv 2023

  5. [4]

    Skarlinski, Sam Cox, Jon M

    Michael D. Skarlinski, Sam Cox, Jon M. Laurent, James D. Braza, Michaela M. Hinks, M. Hammerling, Manvitha Ponnapati, Samuel Rodriques, and Andrew D. 26 White. Language agents achieve superhuman synthesis of scientific knowledge. ArXiv, abs/2409.13740, 2024. URLhttps://api.semanticscholar.org/ CorpusId:272827223

    arXiv 2024

  6. [5]

    Chemlit-qa: a human evaluated dataset for chemistry rag tasks

    Philippe Schwaller. Chemlit-qa: a human evaluated dataset for chemistry rag tasks. MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 6(2), 2025

    2025

  7. [6]

    Benchmarking retrieval-augmented generation for chemistry.arXiv preprint arXiv:2505.07671, 2025

    Xianrui Zhong, Bowen Jin, Siru Ouyang, Yanzhen Shen, Qiao Jin, Yin Fang, Zhiyong Lu, and Jiawei Han. Benchmarking retrieval-augmented generation for chemistry.arXiv preprint arXiv:2505.07671, 2025

    arXiv 2025

  8. [7]

    Rag-enhanced collaborative llm agents for drug discovery.arXiv preprint arXiv:2502.17506, 2025

    Namkyeong Lee, Edward De Brouwer, Ehsan Hajiramezanali, Tommaso Biancalani, Chanyoung Park, and Gabriele Scalia. Rag-enhanced collaborative llm agents for drug discovery.arXiv preprint arXiv:2502.17506, 2025

    arXiv 2025

Show all 123 references
  1. [8]

    Kragen: a knowledge graph-enhanced rag framework for biomedical problem solving using large language models.Bioinfor- matics, 40(6):btae353, 2024

    Nicholas Matsumoto, Jay Moran, Hyunjun Choi, Miguel E Hernandez, Mythreye Venkatesan, Paul Wang, and Jason H Moore. Kragen: a knowledge graph-enhanced rag framework for biomedical problem solving using large language models.Bioinfor- matics, 40(6):btae353, 2024

    2024

  2. [9]

    Fi- nancial report chunking for effective retrieval augmented generation.arXiv preprint arXiv:2402.05131, 2024

    Antonio Jimeno Yepes, Yao You, Jan Milczek, Sebastian Laverde, and Renyu Li. Fi- nancial report chunking for effective retrieval augmented generation.arXiv preprint arXiv:2402.05131, 2024

    2024 arXiv

  3. [10]

    A survey of text representation and embedding techniques in nlp.IEEE Access, 11:36120–36146, 2023

    Rajvardhan Patil, Sorio Boit, Venkat Gudivada, and Jagadeesh Nandigam. A survey of text representation and embedding techniques in nlp.IEEE Access, 11:36120–36146, 2023

    2023

  4. [11]

    Evaluating chunking strategies for retrieval

    Brandon Smith and Anton Troynikov. Evaluating chunking strategies for retrieval. Tech- nical report, Chroma, July 2024. URLhttps://research.trychroma.com/ evaluating-chunking. Chroma Technical Report

    2024

  5. [12]

    Chemteb: Chemical text embedding benchmark, an overview of embedding models performance & efficiency on a specific domain.arXiv preprint arXiv:2412.00532, 2024

    Ali Shiraee Kasmaee, Mohammad Khodadad, Mohammad Arshi Saloot, Nicholas Sherck, Stephen Dokas, Hamidreza Mahyar, and Soheila Samiee. Chemteb: Chemical text embedding benchmark, an overview of embedding models performance & efficiency on a specific domain.arXiv preprint arXiv:2...

    2024 arXiv

  6. [13]

    Natural questions: a benchmark for question answering research.Transactions of the Association for Computational Linguistics, 7:453–466, 2019

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. Natural questions: a benchmark for question answering research.Transactions of the Association for Computat...

    2019

  7. [14]

    Ms marco: A human-generated machine reading comprehension dataset

    Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Ma- jumder, and Li Deng. Ms marco: A human-generated machine reading comprehension dataset. 2016

    2016

  8. [15]

    Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. HotpotQA: A dataset for diverse, ex- plainable multi-hop question answering. In Ellen Riloff, David Chiang, Julia Hock- enmaier, and Jun’ichi Tsujii, editors,P...

    2018 doi

  9. [16]

    Mmteb: Massive multilingual text embedding benchmark.arXiv preprint arXiv:2502.13595, 2025

    Kenneth Enevoldsen, Isaac Chung, Imene Kerboua, Márton Kardos, Ashwin Mathur, David Stap, Jay Gala, Wissam Siblini, Dominik Krzemi ´nski, Genta Indra Winata, Saba Sturua, Saiteja Utpala, Mathieu Ciancone, Marion Schaeffer, Gabriel Sequeira, Di- ganta Misra, Shreeya Dhakal, Jon...

    2025 doi

  10. [17]

    Mteb: Massive text embedding benchmark.arXiv preprint arXiv:2210.07316, 2022

    Niklas Muennighoff, Nouamane Tazi, Loic Magne, and Nils Reimers. Mteb: Massive text embedding benchmark.arXiv preprint arXiv:2210.07316, 2022

    2022 arXiv

  11. [19]

    A comprehensive survey of evaluation techniques for recommendation systems

    Aryan Jadon and Avinash Patil. A comprehensive survey of evaluation techniques for recommendation systems. InInternational Conference on Computation of Artificial In- telligence & Machine Learning, pages 281–304. Springer, 2024. 28

    2024

  12. [20]

    Langchain: Building applications with llms through composability.https://github.com/langchain-ai/langchain,

    Harrison Chase and LangChain contributors. Langchain: Building applications with llms through composability.https://github.com/langchain-ai/langchain,

  13. [21]

    5 levels of text splitting.https://github.com/ FullStackRetrieval-com/RetrievalTutorials/blob/main/ tutorials/LevelsOfTextSplitting/5_Levels_Of_Text_ Splitting.ipynb, 2024

    Greg Kamradt. 5 levels of text splitting.https://github.com/ FullStackRetrieval-com/RetrievalTutorials/blob/main/ tutorials/LevelsOfTextSplitting/5_Levels_Of_Text_ Splitting.ipynb, 2024. Implementation of Semantic Chunking

    2024

  14. [22]

    all-minilm-l6-v2.https://huggingface.co/ sentence-transformers/all-MiniLM-L6-v2, 2021

    Nils Reimers and Iryna Gurevych. all-minilm-l6-v2.https://huggingface.co/ sentence-transformers/all-MiniLM-L6-v2, 2021

    2021

  15. [23]

    Sentence-bert: Sentence embeddings using siamese bert-networks.arXiv preprint arXiv:1908.10084, 2019

    Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks.arXiv preprint arXiv:1908.10084, 2019

    1908 arXiv

  16. [24]

    all-minilm-l12-v2.https://huggingface

    Nils Reimers and Iryna Gurevych. all-minilm-l12-v2.https://huggingface. co/sentence-transformers/all-MiniLM-L12-v2, 2021

    2021

  17. [25]

    all-mpnet-base-v2.https://huggingface

    Nils Reimers and Iryna Gurevych. all-mpnet-base-v2.https://huggingface. co/sentence-transformers/all-mpnet-base-v2, 2021

    2021

  18. [26]

    gtr-t5-base.https://huggingface.co/ sentence-transformers/gtr-t5-base, 2021

    Nils Reimers and Iryna Gurevych. gtr-t5-base.https://huggingface.co/ sentence-transformers/gtr-t5-base, 2021

    2021

  19. [27]

    gtr-t5-large.https://huggingface.co/ sentence-transformers/gtr-t5-large, 2021

    Nils Reimers and Iryna Gurevych. gtr-t5-large.https://huggingface.co/ sentence-transformers/gtr-t5-large, 2021

    2021

  20. [28]

    gtr-t5-xl.https://huggingface.co/ sentence-transformers/gtr-t5-xl, 2021

    Nils Reimers and Iryna Gurevych. gtr-t5-xl.https://huggingface.co/ sentence-transformers/gtr-t5-xl, 2021

    2021

  21. [29]

    bert-base-nli-mean-tokens

    Nils Reimers and Iryna Gurevych. bert-base-nli-mean-tokens. https://huggingface.co/sentence-transformers/ bert-base-nli-mean-tokens, 2019

    2019

  22. [30]

    multi-qa-mpnet-base-dot- v1.https://huggingface.co/sentence-transformers/ multi-qa-mpnet-base-dot-v1, 2021

    Nils Reimers and Iryna Gurevych. multi-qa-mpnet-base-dot- v1.https://huggingface.co/sentence-transformers/ multi-qa-mpnet-base-dot-v1, 2021

    2021

  23. [31]

    paraphrase-multilingual-minilm-l12- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-MiniLM-L12-v2, 2020

    Nils Reimers and Iryna Gurevych. paraphrase-multilingual-minilm-l12- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-MiniLM-L12-v2, 2020

    2020

  24. [32]

    paraphrase-multilingual-mpnet-base- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-mpnet-base-v2, 2021

    Nils Reimers and Iryna Gurevych. paraphrase-multilingual-mpnet-base- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-mpnet-base-v2, 2021. 29

    2021

  25. [33]

    Domain-specific language model pretrain- ing for biomedical natural language processing.ACM Transactions on Computing for Healthcare (HEALTH), 3(1):1–23, 2021

    Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Domain-specific language model pretrain- ing for biomedical natural language processing.ACM Transactions on Computing for Healthcare (HEALTH), 3(1):1–23, 2021

    2021

  26. [34]

    Biomednlp-biomedbert- base-uncased-abstract-fulltext.https://huggingface.co/microsoft/ BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext, 2021

    Yifan Gu, Rajesh Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Biomednlp-biomedbert- base-uncased-abstract-fulltext.https://huggingface.co/microsoft/ BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext, 2021

    2021

  27. [35]

    Biomednlp- biomedbert-base-uncased-abstract.https://huggingface.co/microsoft/ BiomedNLP-BiomedBERT-base-uncased-abstract, 2021

    Yifan Gu, Rajesh Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xi- aodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Biomednlp- biomedbert-base-uncased-abstract.https://huggingface.co/microsoft/ BiomedNLP-BiomedBERT-base-uncased-abstract, 2021

    2021

  28. [36]

    Biomednlp-pubmedbert- base-uncased-abstract-fulltext.https://huggingface.co/microsoft/ BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext, 2021

    Yifan Gu, Rajesh Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Biomednlp-pubmedbert- base-uncased-abstract-fulltext.https://huggingface.co/microsoft/ BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext, 2021

    2021

  29. [37]

    Biomednlp- pubmedbert-base-uncased-abstract.https://huggingface.co/microsoft/ BiomedNLP-PubMedBERT-base-uncased-abstract, 2021

    Yifan Gu, Rajesh Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xi- aodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. Biomednlp- pubmedbert-base-uncased-abstract.https://huggingface.co/microsoft/ BiomedNLP-PubMedBERT-base-uncased-abstract, 2021

    2021

  30. [38]

    Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transform- ers.arXiv preprint arXiv:2002.10957, 2020

    Wenhui Wang, Furu Wei, Li Dong, Hangbo Bao, Nan Yang, and Ming Zhou. Minilm: Deep self-attention distillation for task-agnostic compression of pre-trained transform- ers.arXiv preprint arXiv:2002.10957, 2020

    2002 arXiv

  31. [39]

    Minilm-l12-h384-uncased.https://huggingface.co/microsoft/ MiniLM-L12-H384-uncased, 2020

    Siqi Wang, Furu Wei, Li Dong, Hangbo Bao, Nan Yang, and Ming Zhou. Minilm-l12-h384-uncased.https://huggingface.co/microsoft/ MiniLM-L12-H384-uncased, 2020

    2020

  32. [40]

    Mpnet: Masked and permuted pre-training for language understanding.arXiv preprint arXiv:2004.09297, 2020

    Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. Mpnet: Masked and permuted pre-training for language understanding.arXiv preprint arXiv:2004.09297, 2020

    2004 arXiv

  33. [41]

    mpnet-base.https: //huggingface.co/microsoft/mpnet-base, 2020

    Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. mpnet-base.https: //huggingface.co/microsoft/mpnet-base, 2020

    2020

  34. [42]

    Text embeddings by weakly-supervised contrastive pre- training.arXiv preprint arXiv:2212.03533, 2022

    Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Ran- gan Majumder, and Furu Wei. Text embeddings by weakly-supervised contrastive pre- training.arXiv preprint arXiv:2212.03533, 2022. 30

    2022 arXiv

  35. [43]

    e5-small

    Luyu Wang, Yuwei Zhou, Xiang Ren, Julian McAuley, and Yiming Yang. e5-small. https://huggingface.co/intfloat/e5-small, 2022

    2022

  36. [44]

    e5-small-v2

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. e5-small-v2. https://huggingface.co/intfloat/e5-small-v2, 2022

    2022

  37. [45]

    e5-base.https: //huggingface.co/intfloat/e5-base, 2022

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. e5-base.https: //huggingface.co/intfloat/e5-base, 2022

    2022

  38. [46]

    e5-base-v2

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. e5-base-v2. https://huggingface.co/intfloat/e5-base-v2, 2022

    2022

  39. [47]

    e5-large.https: //huggingface.co/intfloat/e5-large, 2022

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. e5-large.https: //huggingface.co/intfloat/e5-large, 2022

    2022

  40. [48]

    e5-large-v2

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. e5-large-v2. https://huggingface.co/intfloat/e5-large-v2, 2022

    2022

  41. [49]

    multilingual- e5-base.https://huggingface.co/intfloat/multilingual-e5-base, 2022

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. multilingual- e5-base.https://huggingface.co/intfloat/multilingual-e5-base, 2022

    2022

  42. [50]

    multilingual-e5- large.https://huggingface.co/intfloat/multilingual-e5-large, 2022

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. multilingual-e5- large.https://huggingface.co/intfloat/multilingual-e5-large, 2022

    2022

  43. [51]

    multilingual-e5- small.https://huggingface.co/intfloat/multilingual-e5-small, 2022

    Luyu Wang, Yuwei Zhou, Xiang Ren, Jing Zhang, and Yiming Tao. multilingual-e5- small.https://huggingface.co/intfloat/multilingual-e5-small, 2022

    2022

  44. [52]

    C-pack: Packaged resources to advance general chinese embedding, 2023

    Shitao Xiao, Zheng Liu, Peitian Zhang, and Niklas Muennighoff. C-pack: Packaged resources to advance general chinese embedding, 2023

    2023

  45. [53]

    bge-small-en.https://huggingface.co/BAAI/bge-small-en, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-small-en.https://huggingface.co/BAAI/bge-small-en, 2023

    2023

  46. [54]

    bge-small-en-v1.5.https://huggingface.co/BAAI/bge-small-en-v1

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-small-en-v1.5.https://huggingface.co/BAAI/bge-small-en-v1. 5, 2023

    2023

  47. [55]

    bge-base-en.https://huggingface.co/BAAI/bge-base-en, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-base-en.https://huggingface.co/BAAI/bge-base-en, 2023

    2023

  48. [56]

    bge-base-en-v1.5.https://huggingface.co/BAAI/bge-base-en-v1.5, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-base-en-v1.5.https://huggingface.co/BAAI/bge-base-en-v1.5, 2023. 31

    2023

  49. [57]

    bge-large-en.https://huggingface.co/BAAI/bge-large-en, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-large-en.https://huggingface.co/BAAI/bge-large-en, 2023

    2023

  50. [58]

    bge-large-en-v1.5.https://huggingface.co/BAAI/bge-large-en-v1.5, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-large-en-v1.5.https://huggingface.co/BAAI/bge-large-en-v1.5, 2023

    2023

  51. [59]

    Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation, 2024

    Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation, 2024

    2024

  52. [60]

    bge-m3.https://huggingface.co/BAAI/bge-m3, 2023

    Yue Liu, Yixin Wang, Yining Liu, Shuai Wang, Jiaze Xu, Zhiyuan Liu, and Jie Tang. bge-m3.https://huggingface.co/BAAI/bge-m3, 2023

    2023

  53. [61]

    nomic-embed-text-v1.https://huggingface.co/nomic-ai/ nomic-embed-text-v1, 2023

    Nomic AI. nomic-embed-text-v1.https://huggingface.co/nomic-ai/ nomic-embed-text-v1, 2023

    2023

  54. [62]

    Morris, Brandon Duderstadt, and Andriy Mulyar

    Zach Nussbaum, John X. Morris, Brandon Duderstadt, and Andriy Mulyar. Nomic em- bed: Training a reproducible long context text embedder, 2024

    2024

  55. [63]

    nomic-embed-text-v1.5.https://huggingface.co/nomic-ai/ nomic-embed-text-v1.5, 2024

    Nomic AI. nomic-embed-text-v1.5.https://huggingface.co/nomic-ai/ nomic-embed-text-v1.5, 2024

    2024

  56. [64]

    nomic-bert-2048.https://huggingface.co/nomic-ai/ nomic-bert-2048, 2023

    Nomic AI. nomic-bert-2048.https://huggingface.co/nomic-ai/ nomic-bert-2048, 2023

    2023

  57. [65]

    Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    Noam Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    2002 arXiv

  58. [66]

    Unsupervised dense information retrieval with contrastive learning.arXiv preprint arXiv:2112.09118, 2021

    Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. Unsupervised dense information retrieval with contrastive learning.arXiv preprint arXiv:2112.09118, 2021

    2021 arXiv

  59. [67]

    contriever

    Gautier Izacard, Marshall Grusky, Ehsan Hosseini-Asl, and Edouard Grave. contriever. https://huggingface.co/facebook/contriever, 2022

    2022

  60. [68]

    contriever- msmarco.https://huggingface.co/facebook/contriever-msmarco, 2022

    Gautier Izacard, Marshall Grusky, Ehsan Hosseini-Asl, and Edouard Grave. contriever- msmarco.https://huggingface.co/facebook/contriever-msmarco, 2022

    2022

  61. [69]

    Chemberta: Large- scale self-supervised pretraining for molecular property prediction, 2020

    Soham Chithrananda, Gabriel Grand, and Bharath Ramsundar. Chemberta: Large- scale self-supervised pretraining for molecular property prediction, 2020. Available at https://arxiv.org/abs/2010.09885. 32

    2020 arXiv

  62. [70]

    Chemberta-77m-mtr.https://huggingface.co/DeepChem/ ChemBERTa-77M-MTR, 2023

    DeepChem. Chemberta-77m-mtr.https://huggingface.co/DeepChem/ ChemBERTa-77M-MTR, 2023

    2023

  63. [71]

    Bert: Pre- training of deep bidirectional transformers for language understanding.arXiv preprint arXiv:1810.04805, 2018

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre- training of deep bidirectional transformers for language understanding.arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  64. [72]

    jina-embeddings-v2-base-en.https://huggingface.co/jinaai/ jina-embeddings-v2-base-en, 2023

    Jina AI. jina-embeddings-v2-base-en.https://huggingface.co/jinaai/ jina-embeddings-v2-base-en, 2023

    2023

  65. [73]

    Jina embeddings 2: 8192-token general-purpose text embeddings for long documents, 2023

    Michael Günther, Jackmin Ong, Isabelle Mohr, Alaeddine Abdessalem, Tanguy Abel, Mohammad Kalim Akram, Susana Guzman, Georgios Mastrapas, Saba Sturua, Bo Wang, Maximilian Werk, Nan Wang, and Han Xiao. Jina embeddings 2: 8192-token general-purpose text embeddings for long docume...

    2023

  66. [74]

    jina-embeddings-v2-small-en.https://huggingface.co/jinaai/ jina-embeddings-v2-small-en, 2023

    Jina AI. jina-embeddings-v2-small-en.https://huggingface.co/jinaai/ jina-embeddings-v2-small-en, 2023

    2023

  67. [75]

    Matscibert: A materials domain lan- guage model.https://huggingface.co/m3rg-iitd/matscibert, 2021

    Danish Rana, Alok Choudhary, and Ankit Agrawal. Matscibert: A materials domain lan- guage model.https://huggingface.co/m3rg-iitd/matscibert, 2021

    2021

  68. [77]

    Gemini: General embeddings from mixed-initiative instruction.https: //huggingface.co/describeai/gemini, 2024

    DescribeAI. Gemini: General embeddings from mixed-initiative instruction.https: //huggingface.co/describeai/gemini, 2024

    2024

  69. [78]

    Instructor-xl: Teaching models with instruction and feedback.https://huggingface.co/hkunlp/ instructor-xl, 2022

    Hanzi Wang, Yujia Zhou, Xiang Liu, and Minlie Huang. Instructor-xl: Teaching models with instruction and feedback.https://huggingface.co/hkunlp/ instructor-xl, 2022

    2022

  70. [79]

    One embedder, any task: Instruction-finetuned text embeddings.arXiv preprint arXiv:2212.09741, 2022

    Hongjin Su, Weijia Shi, Jungo Kasai, Yizhong Wang, Yushi Hu, Mari Ostendorf, Wen- tau Yih, Noah A Smith, Luke Zettlemoyer, and Tao Yu. One embedder, any task: Instruction-finetuned text embeddings.arXiv preprint arXiv:2212.09741, 2022

    2022 arXiv

  71. [80]

    Scibert: A pretrained language model for scientific text.arXiv preprint arXiv:1903.10676, 2019

    Iz Beltagy, Kyle Lo, and Arman Cohan. Scibert: A pretrained language model for scientific text.arXiv preprint arXiv:1903.10676, 2019

    1903 arXiv

  72. [81]

    Scibert: A pretrained language model for scientific text

    Iz Beltagy, Kyle Lo, and Arman Cohan. Scibert: A pretrained language model for scientific text. InEMNLP. Association for Computational Linguistics, 2019. URL https://www.aclweb.org/anthology/D19-1371. 33

    2019

  73. [82]

    Specter: Document-level representation learning using citation-informed transformers.ACL, 2020

    Arman Cohan, Sergey Feldman, Iz Beltagy, Doug Downey, and Daniel Weld. Specter: Document-level representation learning using citation-informed transformers.ACL, 2020

    2020

  74. [83]

    allenai-specter.https://huggingface.co/sentence-transformers/ allenai-specter, 2020

    Arman Cohan, Sergey Feldman, Iz Beltagy, Doug Downey, and Daniel S Weld. allenai-specter.https://huggingface.co/sentence-transformers/ allenai-specter, 2020

    2020

  75. [84]

    Chemicalbert: Pretrained transformers for chemistry.https:// huggingface.co/recobo/chemical-bert-uncased, 2023

    Recobo AI. Chemicalbert: Pretrained transformers for chemistry.https:// huggingface.co/recobo/chemical-bert-uncased, 2023

    2023

  76. [85]

    paraphrase-multilingual-mpnet-base- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-mpnet-base-v2, 2021

    Nils Reimers and Iryna Gurevych. paraphrase-multilingual-mpnet-base- v2.https://huggingface.co/sentence-transformers/ paraphrase-multilingual-mpnet-base-v2, 2021

    2021

  77. [86]

    New embedding models and api updates

    OpenAI. New embedding models and api updates. Technical re- port, OpenAI, 2024. URLhttps://openai.com/index/ new-embedding-models-and-api-updates/. Accessed: 2025-05-19

    2024

  78. [87]

    Recent advances in text embedding: A comprehensive review of top- performing methods on the mteb benchmark.arXiv preprint arXiv:2406.01607, 2024

    Hongliu Cao. Recent advances in text embedding: A comprehensive review of top- performing methods on the mteb benchmark.arXiv preprint arXiv:2406.01607, 2024

    2024 arXiv

  79. [88]

    Multi-paragraph segmentation of expository text.arXiv preprint cmp- lg/9406037, 1994

    Marti A Hearst. Multi-paragraph segmentation of expository text.arXiv preprint cmp- lg/9406037, 1994

    1994

  80. [89]

    Linear text segmentation: approaches, advances and applications.Proc

    Freddy Choi. Linear text segmentation: approaches, advances and applications.Proc. of CLUK3, 2000

    2000

  81. [90]

    A statistical model for domain-independent text segmentation

    Masao Utiyama and Hitoshi Isahara. A statistical model for domain-independent text segmentation. InProceedings of the 39th annual meeting of the Association for Compu- tational Linguistics, pages 499–506, 2001

    2001

  82. [91]

    Modeling local coherence: An entity-based ap- proach.Computational Linguistics, 34(1):1–34, 2008

    Regina Barzilay and Mirella Lapata. Modeling local coherence: An entity-based ap- proach.Computational Linguistics, 34(1):1–34, 2008

    2008

  83. [92]

    Text segmentation: A topic modeling perspective.Information Processing & Management, 47(4):528–544, 2011

    Hemant Misra, Franccois Yvon, Olivier Cappe, and Joemon Jose. Text segmentation: A topic modeling perspective.Information Processing & Management, 47(4):528–544, 2011

    2011

  84. [93]

    Text segmentation with topic models.Journal for Language Technology and Computational Linguistics, 27(1):47–69, 2012

    Martin Riedl and Chris Biemann. Text segmentation with topic models.Journal for Language Technology and Computational Linguistics, 27(1):47–69, 2012

    2012

  85. [94]

    Unsupervised text segmen- tation using semantic relatedness graphs

    Goran Glavavs, Federico Nanni, and Simone Paolo Ponzetto. Unsupervised text segmen- tation using semantic relatedness graphs. InProceedings of the Fifth Joint Conference on 34 Lexical and Computational Semantics, pages 125–130. Association for Computational Linguistics, 2016

    2016

  86. [95]

    Unsupervised topic segmentation of meetings with bert embed- dings.arXiv preprint arXiv:2106.12978, 2021

    Alessandro Solbiati, Kevin Heffernan, Georgios Damaskinos, Shivani Poddar, Shubham Modi, and Jacques Cali. Unsupervised topic segmentation of meetings with bert embed- dings.arXiv preprint arXiv:2106.12978, 2021

    2021 arXiv

  87. [96]

    Uncovering the potential of chatgpt for discourse analysis in dialogue: An empirical study.arXiv preprint arXiv:2305.08391, 2023

    Yaxin Fan, Feng Jiang, Peifeng Li, and Haizhou Li. Uncovering the potential of chatgpt for discourse analysis in dialogue: An empirical study.arXiv preprint arXiv:2305.08391, 2023

    2023 arXiv

  88. [97]

    Super- dialseg: A large-scale dataset for supervised dialogue segmentation.arXiv preprint arXiv:2305.08371, 2023

    Junfeng Jiang, Chengzhang Dong, Sadao Kurohashi, and Akiko Aizawa. Super- dialseg: A large-scale dataset for supervised dialogue segmentation.arXiv preprint arXiv:2305.08371, 2023

    2023 arXiv

  89. [98]

    Text segmentation as a supervised learning task.arXiv preprint arXiv:1803.09337, 2018

    Omri Koshorek, Adir Cohen, Noam Mor, Michael Rotman, and Jonathan Berant. Text segmentation as a supervised learning task.arXiv preprint arXiv:1803.09337, 2018

    2018 arXiv

  90. [99]

    Text seg- mentation by cross segment attention.arXiv preprint arXiv:2004.14535, 2020

    Michal Lukasik, Boris Dadachev, Gonccalo Simoes, and Kishore Papineni. Text seg- mentation by cross segment attention.arXiv preprint arXiv:2004.14535, 2020

    2004 arXiv

  91. [100]

    Im- proving long document topic segmentation models with enhanced coherence modeling

    Hai Yu, Chong Deng, Qinglin Zhang, Jiaqing Liu, Qian Chen, and Wen Wang. Im- proving long document topic segmentation models with enhanced coherence modeling. arXiv preprint arXiv:2310.11772, 2023

    2023 arXiv

  92. [101]

    Transformer over pre-trained transformer for neural text segmentation with enhanced topic coherence

    Kelvin Lo, Yuan Jin, Weicong Tan, Ming Liu, Lan Du, and Wray Buntine. Transformer over pre-trained transformer for neural text segmentation with enhanced topic coherence. arXiv preprint arXiv:2110.07160, 2021

    2021 arXiv

  93. [102]

    A joint model for document segmentation and segment labeling

    Joe Barrow, Rajiv Jain, Vlad Morariu, Varun Manjunatha, Douglas W Oard, and Philip Resnik. A joint model for document segmentation and segment labeling. InProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 313– 322, 2020

    2020

  94. [103]

    Two-level transformer and auxiliary coherence modeling for improved text segmentation

    Swapna Somasundaran et al. Two-level transformer and auxiliary coherence modeling for improved text segmentation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 7797–7804, 2020

    2020

  95. [104]

    Tipster: A topic-guided language model for topic-aware text segmentation

    Zheng Gong, Shiwei Tong, Han Wu, Qi Liu, Hanqing Tao, Wei Huang, and Runlong Yu. Tipster: A topic-guided language model for topic-aware text segmentation. In International Conference on Database Systems for Advanced Applications, pages 213–

  96. [105]

    Recent trends in linear text segmentation: A survey.arXiv preprint arXiv:2411.16613, 2024

    Iacopo Ghinassi, Lin Wang, Chris Newell, and Matthew Purver. Recent trends in linear text segmentation: A survey.arXiv preprint arXiv:2411.16613, 2024

    2024 arXiv

  97. [106]

    Text clustering with llm embeddings.arXiv e-prints, pages arXiv–2403, 2024

    Alina Petukhova, Joao P Matos-Carvalho, and Nuno Fachada. Text clustering with llm embeddings.arXiv e-prints, pages arXiv–2403, 2024

    2024

  98. [107]

    Efficient estimation of word representations in vector space.arXiv preprint arXiv:1301.3781, 2013

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space.arXiv preprint arXiv:1301.3781, 2013

    2013 arXiv

  99. [108]

    Glove: Global vectors for word representation

    Jeffrey Pennington, Richard Socher, and Christopher D Manning. Glove: Global vectors for word representation. InProceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532–1543, 2014

    2014

  100. [109]

    Enriching word vectors with subword information.Transactions of the association for computational linguistics, 5:135–146, 2017

    Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. Enriching word vectors with subword information.Transactions of the association for computational linguistics, 5:135–146, 2017

    2017

  101. [110]

    Study of extractive text summarizer using the elmo embedding

    Hritvik Gupta and Mayank Patel. Study of extractive text summarizer using the elmo embedding. In2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC), pages 829–834. IEEE, 2020

    2020

  102. [111]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. InProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human langua...

    2019

  103. [112]

    Improving language understanding by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language understanding by generative pre-training. 2018

    2018

  104. [113]

    Towards general text embeddings with multi-stage contrastive learning.arXiv preprint arXiv:2308.03281, 2023

    Zehan Li, Xin Zhang, Yanzhao Zhang, Dingkun Long, Pengjun Xie, and Meishan Zhang. Towards general text embeddings with multi-stage contrastive learning.arXiv preprint arXiv:2308.03281, 2023

    2023 arXiv

  105. [114]

    C-pack: Packed resources for general chinese embeddings

    Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. C-pack: Packed resources for general chinese embeddings. InProceedings of the 47th international ACM SIGIR conference on research and development in information retrieval, pages 641–649, 2024

    2024

  106. [115]

    Sentence-transformers.https://www.sbert

    Nils Reimers and Iryna Gurevych. Sentence-transformers.https://www.sbert. net, 2019

    2019

  107. [116]

    Domain-specific language model 36 pretraining for biomedical natural language processing, 2020

    Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tris- tan Naumann, Jianfeng Gao, and Hoifung Poon. Domain-specific language model 36 pretraining for biomedical natural language processing, 2020. Available athttps: //arxiv.org/abs/2007.15779

    2020 arXiv

  108. [117]

    Biomedbert: A pre-trained biomedical language model for qa and ir, 2020

    Souradip Chakraborty, Ekaba Bisong, Shweta Bhatt, Thomas Wagner, Riley Elliott, and Francesco Mosconi. Biomedbert: A pre-trained biomedical language model for qa and ir, 2020. URLhttps://aclanthology.org/2020.coling-main.59/

    2020

  109. [118]

    Chemicalbert: A bert model for the chemical domain, 2021

    Recobo. Chemicalbert: A bert model for the chemical domain, 2021. Available at https://huggingface.co/recobo/chemical-bert-uncased

    2021

  110. [119]

    Tanishq Gupta, Mohd Zaki, N. M. Anoop Krishnan, and Mausam. Matscibert: A materials domain language model for text mining and information extraction.npj Computational Materials, 8(1):102, 2022. doi: 10.1038/s41524-022-00784-w. URL https://www.nature.com/articles/s41524-022-00784-w

    2022 doi

  111. [121]

    Jina embeddings v3: Multilingual embeddings with task lora

    Saba Sturua, Isabelle Mohr, Mohammad Kalim Akram, Michael Gunther, Bo Wang, Markus Krimmel, Feng Wang, Georgios Mastrapas, Andreas Koukounas, Nan Wang, and Han Xiao. Jina embeddings v3: Multilingual embeddings with task lora. arXiv preprint arXiv:2409.10173, 2024. URLhttps://a...

    2024 arXiv

  112. [122]

    Multilingual e5 text embeddings: A technical report.arXiv preprint arXiv:2402.05672, 2024

    Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. Multilingual e5 text embeddings: A technical report.arXiv preprint arXiv:2402.05672, 2024. URLhttps://arxiv.org/abs/2402.05672

    2024 arXiv

  113. [123]

    Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity embedding model.arXiv preprint arXiv:2402.03216, 2024

    Yiming Xu, Yuxuan Liu, Yichang Zhang, Xinyu Wang, Ziyang Liu, Yiming Wang, Yu- jie Wang, Yuxuan Wang, Xiangyang Zhang, Zhiyuan Liu, et al. Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity embedding model.arXiv preprint arXiv:2402.03216, 2024. URLhttps://...

    2024 arXiv

  114. [2022]

    Accessed: 2025-05-28

    2025

  115. [2024]

    URLhttps://arxiv.org/abs/2502.07972

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.