Pith. sign in

REVIEW 4 major objections 6 minor 25 references

Evaluating the Robustness of Dense Retrievers in Interdisciplinary Domains

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that benchmark topic structure controls how much benefit domain adaptation appears to give a dense retriever: overlapping topics yield up to a 3.6-fold larger NDCG gain than well-separated topics.

desk verdict A clean case study showing benchmark choice flips adaptation conclusions, but the headline 3.6x causal claim is confounded by structural differences between the two benchmarks. read the letter →

arxiv 2506.21581 v1 pith:H43G6LL3 submitted 2025-06-16 cs.IR cs.AIcs.LG

Sarthak Chaturvedi , Anurag Acharya , Rounak Meyur , Koby Hayashi , Sai Munikoti , Sameera Horawalavithana This is my paper · ORCID
classification cs.IRcs.AIcs.LG
keywords denseretrievaldomainadaptationevaluationbenchmarkdesigntopicdiversityboundaryoverlapNDCGenvironmentalimpactstatements
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 tries to establish that evaluation benchmarks themselves, not just the model, determine how valuable domain adaptation looks. Using environmental regulatory documents (Environmental Impact Statements) as a case study, the authors fine-tune a late-interaction dense retriever on synthetic question-passage pairs and test it on two benchmarks drawn from similar federal agencies. On the benchmark with clearly separated topic clusters, the best fine-tuned model gains at most 0.61% NDCG over the base model; on the benchmark with overlapping topics, the same approach gains up to 2.22% NDCG, a 3.6-fold difference. The paper concludes that benchmarks with well-separated topics systematically underestimate adaptation benefits, and that topic diversity metrics such as cosine distance and silhouette score track this difference. If right, this matters for any high-stakes deployment decision that relies on retrieval benchmarks to decide whether fine-tuning is worth it.

What carries the argument

The load-bearing mechanism is the semantic structure of the evaluation benchmark, operationalized by three topic-diversity measures: average cosine distance between context embeddings (semantic spread), silhouette score (how cleanly contexts cluster into topics), and topic entropy (uniformity of the topic distribution). The paper uses these metrics to characterize NEPAQuAD-LLM as having greater topic overlap than NEPAQuAD-SME-LLM, then connects that contrast to the observed 3.6-fold difference in NDCG gain. The evaluation procedure itself—ranking all contexts per question and scoring with NDCG@89 or NDCG@507—is what converts identical fine-tuned models into different apparent improvements.

What would settle it

Use a single EIS corpus and construct two test sets with identical size, question types, and baseline NDCG, differing only in how much topic clusters overlap (for example, by mixing sentences from different topics into the same passage); measure NDCG gains from the same fine-tuned models, and if gains do not increase as overlap increases, the central claim is refuted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that identical domain adaptation procedures produce benchmark-dependent conclusions: ColBERTv2 fine-tuned on 10, 100, or 700 EIS documents shows small NDCG gains (maximum 0.61%, with the 100-document fine-tune) on NEPAQuAD-SME-LLM, a benchmark whose 89 contexts form distinct topic clusters, and larger gains (up to 2.22%, with the 700-document fine-tune) on NEPAQuAD-LLM, whose 507 contexts overlap semantically. The paper attributes the discrepancy to topic-boundary characteristics, quantifying them as 11.1% higher average cosine distance and 23.2% lower silhouette score for NEPAQuAD-LLM. Because both benchmarks draw on EIS documents from similar federal agencies, the authors argue the difference stems from topic structure rather than document source or agency terminology, and conclude that evaluation frameworks with well-separated topics regularly underestimate domain adaptation benefits.

Load-bearing premise

The whole conclusion rests on the assumption that the two benchmarks differ only in how much their topics overlap, when in fact they also differ in size, in how many questions share each passage, and in how hard the retrieval task is to begin with.

Editorial extensions

If this is right

  • A benchmark whose topic clusters are well separated can make a fine-tuned retriever look almost unchanged (max +0.61% NDCG), while the same fine-tuning on an overlapping-topic benchmark looks clearly beneficial (up to +2.22% NDCG).
  • Topic-diversity statistics—cosine distance, silhouette score, and topic entropy—are useful diagnostics for whether an evaluation benchmark will be sensitive to domain adaptation.
  • For high-stakes regulatory search, the extra NDCG gain at top ranks is the difference between surfacing and missing relevant requirements such as 'habitat restoration' versus 'habitat mitigation'.
  • Reported gains from domain adaptation should be read as properties of the model–benchmark pair, not of the model alone; evaluation frameworks with well-separated topics will regularly understate the value of adaptation.

Reading between the lines

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

  • Editorial extension: because the two benchmarks differ in more than topic overlap (89 vs 507 contexts, about 18 vs 1 questions per context, baseline NDCG 0.9749 vs 0.8902), the clean test of the paper's mechanism is a benchmark continuum that varies only topic overlap while holding those factors fixed.
  • Editorial extension: such a continuum could be built by starting from one EIS corpus and gradually mixing sentences from different topics into passages, then checking whether NDCG gains rise monotonically as silhouette score falls.
  • Editorial extension: on the paper's own numbers the best model differs by benchmark—the 100-document fine-tune on the well-separated set, the 700-document fine-tune on the overlapping set—so single-benchmark leaderboards can reverse the apparent winner of domain adaptation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 studies how evaluation benchmark characteristics influence the measured benefit of domain adaptation for dense retrieval. Using environmental regulatory documents (EISs) as a case study, the authors fine-tune ColBERTv2 on synthetic question-context pairs generated from 10, 100, or 700 EIS documents by Gemini 1.5 Pro, and evaluate on two benchmarks: NQ-SME-LLM (1,589 pairs over 89 unique contexts) and NQ-LLM (556 pairs over 507 unique contexts). They report maximum NDCG gains of 0.61% on NQ-SME-LLM and 2.22% on NQ-LLM, a 3.6-fold difference, and attribute this difference to topic boundary overlap, supported by descriptive metrics showing that NQ-LLM has 11% higher average cosine distance and 23% lower silhouette score. The paper concludes that benchmarks with well-separated topics regularly underestimate domain adaptation benefits.

Significance. The research question is important: evaluation methodology can plausibly change conclusions about whether domain adaptation helps. The design of varying training-data scale (10/100/700 documents) is a strength, as is the descriptive topic-diversity analysis in Table 2 and Figure 1. However, the central causal claim is not supported by the current comparison because the two benchmarks differ simultaneously in corpus size, questions-per-context ratio, baseline headroom, and generation procedure. As it stands, the paper is a cautionary anecdote about two benchmarks rather than a demonstrated relationship between topic boundary overlap and measured adaptation gain; the overgeneralized abstract and conclusion go beyond the evidence.

major comments (4)
  1. [Section 3.4.1 / Table 3 / Section 4.2] The central 3.6-fold difference is confounded by structural differences between the two benchmarks. NQ-SME-LLM has 89 unique contexts and a baseline NDCG of 0.9749, leaving at most about 2.5 points of NDCG headroom, while NQ-LLM has 507 contexts and a baseline of 0.8902, leaving about 11 points of headroom. The benchmarks also differ in questions-per-context (about 18 vs. about 1) and in the NDCG cutoff used (89 vs. 507). The observed gain difference (0.61% vs. 2.22%) is fully consistent with ceiling effects and rank-pool size effects, independent of topic overlap. The 'source similarity' control in Section 4.3 matches only agency distribution; it does not control for corpus size, headroom, or questions-per-context. To support the claim, the authors would need matched evaluations, for example by subsampling NQ-LLM to 89 contexts, matching questions-per-context, or comparing at a common NDCG@k.
  2. [Section 3.2.1 / Appendix E] There is a potential circularity in the NQ-LLM evaluation. The training question-context pairs are generated by Gemini 1.5 Pro using a prompt that asks for 'high-quality, diverse synthetic dataset' questions from document chunks, while the NQ-LLM benchmark is generated by the same model using a closely related prompt with the same six question types (inference, closed-ended, comparison, process, divergent, evaluation). Fine-tuned models are therefore evaluated on questions drawn from the same synthetic distribution used for training, so gains on NQ-LLM may reflect distributional alignment to the generator rather than adaptation to the NEPA domain. The paper should address this by evaluating on human-written queries, by generating the benchmark with a different model or a different prompt, or by quantifying the overlap between training and evaluation question distributions.
  3. [Abstract / Section 4.2 / Section 5] The generalization that 'evaluation frameworks with well-separated topics regularly underestimate domain adaptation benefits' is not supported by two benchmarks, a point the Limitations section itself concedes ('we examined only two benchmarks'). The abstract and conclusion state a general regularity that the uncontrolled comparison cannot establish. The claim should be restricted to these two datasets unless additional benchmarks with intermediate levels of topic overlap are added and show a monotonic relationship.
  4. [Table 2 / Figure 1 / Section 4.2] The topic-diversity metrics are descriptive aggregate statistics and are not connected to query-level retrieval outcomes. No statistical test, confidence interval, or error bar is given for the 3.6-fold gain difference, and the relationship between silhouette score, cosine distance, and adaptation benefit is asserted rather than demonstrated. A query-level analysis that stratifies questions by local topic overlap or by baseline headroom would allow a direct test of whether topic boundary characteristics, rather than other benchmark properties, predict adaptation gains.
minor comments (6)
  1. [Section 4.2] The text states that all models achieved 'NDCG@5 > 0.97' for NQ-SME-LLM, but Section 3.4.4 and Table 3 report NDCG@89 for that benchmark; the cutoff is inconsistent and should be corrected.
  2. [Figure 2] The caption says 'NEPAQuAD-LLM (k=509)', but Section 3.4.1 reports 507 unique contexts for NQ-LLM; the number is inconsistent.
  3. [Section 1] The first sentence refers to the 'National Environment Policy Act', but the correct name is the 'National Environmental Policy Act' (NEPA).
  4. [Appendix A.1 / Figure 4] Figure 4 uses the label 'Nepa-Quad' instead of NEPAQuAD/NQ-LLM, and the caption lists k values [1,5,10,20,30,80] while the body text refers to NDCG@89 and NDCG@507; the notation should be unified.
  5. [References] The reference list contains duplicate entries: ColBERTv2 appears as [15] and [16], and BEIR appears as [17] and [18]; these should be consolidated.
  6. [Section 3.2.1] The text says 'An example prompt is shown above' but the prompt appears below the paragraph; the text and layout should be reordered for clarity.

Circularity Check

1 steps flagged · score 5.0 of 10

NQ-LLM is generated by the same Gemini 1.5 Pro system with a near-identical prompt as the synthetic training data, so the headline 3.6x adaptation-gain difference is partly an artifact of train/eval distributional overlap.

  1. fitted input called prediction [Section 3.2.1 (training prompt) vs Appendix E (NQ-LLM generation prompt); evaluation benchmark defined in Section 3.4.1]
    "Training prompt: "You are an expert AI assisting in creating a high-quality, diverse synthetic dataset to train information retrieval models. Analyze the following document chunk and generate potential queries..." Appendix E prompt: "You are an advanced AI system... Your task is to assist in creating a high-quality, diverse synthetic dataset for training information retrieval models.""

    ColBERTv2 is fine-tuned on question-context pairs generated by Gemini 1.5 Pro (Section 3.2.1) and then evaluated on NQ-LLM, which is also generated by Gemini 1.5 Pro using a near-identical 'create a high-quality, diverse synthetic dataset for training information retrieval models' instruction (Appendix E). NQ-LLM is therefore a sample from the same synthetic query distribution used for training, not an independent probe of the NEPA domain. The 2.22% NDCG gain on NQ-LLM (vs 0.61% on NQ-SME-LLM, which includes SME input and is not generated by the same pipeline) is partly a measure of the model's fit to the generator's question style, so the claimed 3.6x difference is partially forced by construction rather than by topic-boundary overlap.

full rationale

The paper's central experimental chain is otherwise self-contained: topic-diversity metrics (silhouette score, cosine distance, topic entropy) are computed from the benchmark contexts themselves, and NDCG values are obtained from actual rankings, so those numbers are not equivalent to inputs by definition. The self-citation [12] for NQ-SME-LLM is not load-bearing because the paper independently quantifies topic structure and reports full performance tables. However, the load-bearing comparison between NQ-LLM and NQ-SME-LLM is compromised: the two benchmarks differ in corpus size (507 vs 89 unique contexts), questions-per-context (556 vs 1589 pairs), and baseline NDCG (0.8902 vs 0.9749), none of which Section 4.3's 'source similarity' control addresses. More importantly, NQ-LLM's questions are produced by the same Gemini 1.5 Pro model with a prompt nearly identical to the one used to create the fine-tuning data, so gains on NQ-LLM reflect generator alignment as much as domain adaptation. The Limitations section only concedes 'only two benchmarks' and 'synthetic questions generated by large language models,' not this overlap, and the abstract's 'regularly underestimate' generalization is not supported by the uncontrolled two-benchmark comparison. These are validity threats; the train/eval generator overlap is the one that rises to partial circularity because the evaluation distribution is constructed from the same generator as the training distribution.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

No free parameters are fitted to the outcome except the evaluation cutoff choice; the main load-bearing assumptions are the comparability of the two benchmarks and the validity of single-gold relevance labels. The paper introduces no new entities.

free parameters (1)
  • NDCG cutoff k = 89 for NQ-SME-LLM, 507 for NQ-LLM
    Chosen to equal the number of unique contexts in each benchmark, which changes how much headroom exists for improvement and makes the metric not comparable across the two benchmarks. This choice is load-bearing for the reported 3.6x difference.
assumptions (4)
  • domain assumption A single gold-standard context (the one the LLM used to generate the question) is the only relevant context for each query; all other contexts are irrelevant.
    Stated in Section 3.4.3. In a corpus with 507 contexts, many contexts may contain information that answers the question, making this binary relevance assumption questionable and potentially penalizing correct alternate answers.
  • ad hoc to paper The observed NDCG difference between benchmarks is attributable to topic diversity and boundary overlap, not to differences in corpus size, question-per-context ratio, or baseline headroom.
    Invoked in Section 4.3 to interpret the results, but never tested; the benchmarks differ on all these dimensions simultaneously.
  • domain assumption Synthetic questions generated by Gemini 1.5 Pro reflect the information needs of regulatory practitioners.
    The paper's own Limitations section concedes that synthetic questions 'may not fully capture the specific information needs of actual regulatory practitioners.'
  • domain assumption NDCG at the full corpus size is a valid comparison metric across benchmarks with different numbers of candidates.
    The paper reports NDCG@89 and NDCG@507 (Section 3.4.4). These are different ranking problems, so the metric values are not directly comparable without correcting for corpus size.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating the Robustness of Dense Retrievers in Interdisciplinary Domains." pith.science (2026). https://pith.science/paper/H43G6LL3

@misc{pith2026250621581,
  author       = {Pith},
  title        = {Pith review of: Evaluating the Robustness of Dense Retrievers in Interdisciplinary Domains},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H43G6LL3}},
  note         = {Machine review of arXiv:2506.21581}
}
read the original abstract

Evaluation benchmark characteristics may distort the true benefits of domain adaptation in retrieval models. This creates misleading assessments that influence deployment decisions in specialized domains. We show that two benchmarks with drastically different features such as topic diversity, boundary overlap, and semantic complexity can influence the perceived benefits of fine-tuning. Using environmental regulatory document retrieval as a case study, we fine-tune ColBERTv2 model on Environmental Impact Statements (EIS) from federal agencies. We evaluate these models across two benchmarks with different semantic structures. Our findings reveal that identical domain adaptation approaches show very different perceived benefits depending on evaluation methodology. On one benchmark, with clearly separated topic boundaries, domain adaptation shows small improvements (maximum 0.61% NDCG gain). However, on the other benchmark with overlapping semantic structures, the same models demonstrate large improvements (up to 2.22% NDCG gain), a 3.6-fold difference in the performance benefit. We compare these benchmarks through topic diversity metrics, finding that the higher-performing benchmark shows 11% higher average cosine distances between contexts and 23% lower silhouette scores, directly contributing to the observed performance difference. These results demonstrate that benchmark selection strongly determines assessments of retrieval system effectiveness in specialized domains. Evaluation frameworks with well-separated topics regularly underestimate domain adaptation benefits, while those with overlapping semantic boundaries reveal improvements that better reflect real-world regulatory document complexity. Our findings have important implications for developing and deploying AI systems for interdisciplinary domains that integrate multiple topics.

Figures

Figures reproduced from arXiv: 2506.21581 by the authors.

Figure 1
Figure 1. Topic boundary visualization comparing NEPAQuAD-SME-LLM (a) and NEPAQuAD-LLM (b). NEPAQuAD-SME-LLM [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. NDCG improvement comparison across benchmarks. NEPAQuAD-LLM ( [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. NDCG@k performance comparison of ColBERTv2 models on NEPAQuAD-LLM. All models show performance [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Gain in Mean NDCG for different values of [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Model Performance by Document Type • Early Stopping: Implemented early stopping criteria based on validation loss to prevent overfitting and promote gen￾eralization. • Consistency Across Experiments: Maintained the same hy￾perparameters and training configurations acro…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 19 canonical work pages

  1. [1]

    Berkin Alkan, Bekir Bilgehan Tekin, Alper Karamanlıoğlu, and İsmail Karakaya

    work page

  2. [2]

    Fergadiotis, P

    Ilias Chalkidis, M. Fergadiotis, P. Malakasiotis, N. Aletras, and I. Androutsopoulos

    work page

  3. [3]

    Benjamin Clavié, Omar Khattab, Harrison Chase, Anirudh Dharmarajan, Josh Purtell, Minh Nguyen, PrimoUomo89, tm17 abcgen, Diego Peláez Paquico, Jo- hannes Aalto, Patrick, Peter Goldstein, Shaurya Rohatgi, Théo Q., Vishal Bakshi, corrius, mauryaland, Sami, Jan Luca Scheerer, James, Géraud Bourdin, German Martin, Gautam, Deven Mistry, Dale Hille, and Alex Pe...

    work page 2025

  4. [4]

    Abe Bohan Hou, Orion Weller, Guanghui Qin, Eugene Yang, Dawn Lawrie, Nils Holzenberger, Andrew Blair-Stanek, and Benjamin Van Durme. 2024. CLERC: A Dataset for Legal Case Retrieval and Retrieval-Augmented Analysis Generation. arXiv:2406.17186 [cs.CL] https://arxiv.org/abs/2406.17186

    arXiv 2024

  5. [5]

    Wang, and Graham Neubig

    Jennifer Hsia, Afreen Shaikh, Zhiruo. Wang, and Graham Neubig. 2024. RAGGED: Towards Informed Design of Retrieval Augmented Generation Systems. In NeurIPS Workshop on Adaptive Foundation Models

    work page 2024

  6. [6]

    Järvelin and J

    K. Järvelin and J. Kekäläinen. 2002. Cumulated gain-based evaluation of IR techniques.ACM Transactions on Information Systems (TOIS)20, 4 (2002), 422–

    work page 2002

  7. [7]

    Omar Khattab and Matei Zaharia. 2020. ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 39–48

    2020

  8. [8]

    Dayoon Ko, Jinyoung Kim, Sohyeon Kim, Jinhyuk Kim, Jaehoon Lee, Seonghak Song, Minyoung Lee, and Gunhee Kim. 2025. When Should Dense Retrievers Be Updated in Evolving Corpora? Detecting Out-of-Distribution Corpora Using GradNormIR. arXiv:2506.01877 [cs.IR] https://arxiv.org/abs/2506.01877

    work page Pith review arXiv 2025

Show all 25 references
  1. [9]

    Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chang- Hwan So, and Jaewoo Kang. 2020. BioBERT: a Pre-trained Biomedical Language Representation Model for Biomedical Text Mining.Bioinformatics36, 4 (2020), 1234–1240

    2020

  2. [10]

    Alistair Moffat, Paul Bailey, Fritz Scholer, and Peter Thomas. 2017. Incorporating User Expectations and Behavior into the Measurement of Search Effectiveness. ACM Transactions on Information Systems35, 3 (2017), 1–38

    2017

  3. [11]

    Ani Nenkova, Kathleen McKeown, Cristina Rosé, and Julia Hirschberg. 2010. A Framework for Assessing Information Quality and Trustworthiness of Digital Information Sources. InProceedings of the 48th Annual Meeting of the Association for Computational Linguistics. 51–55

    2010

  4. [12]

    Hung Phan, Anurag Acharya, Rounak Meyur, Sarthak Chaturvedi, Shivam Sharma, Mike Parker, Dan Nally, Ali Jannesari, Karl Pazdernik, Mahantesh Ha- lappanavar, Sai Munikoti, and Sameera Horawalavithana. 2024. Examining Long-Context Large Language Models for Environmental Review D...

    2024 arXiv

  5. [13]

    Zobel, and Alistair Moffat

    Lida Rashidi, J. Zobel, and Alistair Moffat. 2021. Evaluating the Predictivity of IR Experiments. doi:10.1145/3404835.3463040

    2021

  6. [14]

    Jon Saad-Falcon, Omar Khattab, Keshav Santhanam, Radu Florian, Martin Franz, Salim Roukos, Avirup Sil, Md Arafat Sultan, and Christopher Potts. 2023. UDAPDR: unsupervised domain adaptation via LLM prompting and distillation of rerankers.arXiv preprint arXiv:2303.00807(2023)

    2023 arXiv

  7. [15]

    Santhanam, Omar Khattab, and Christopher Ré

    K. Santhanam, Omar Khattab, and Christopher Ré. 2021. ColBERTv2: Effec- tive and Efficient Retrieval via Lightweight Late Interaction.arXiv preprint arXiv:2112.01488(2021)

    2021 arXiv

  8. [16]

    Khattab, Jon Saad-Falcon, Christopher Potts, and M

    Keshav Santhanam, O. Khattab, Jon Saad-Falcon, Christopher Potts, and M. Zaharia. 2021. ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction. 3715-3734 pages. doi:10.18653/v1/2022.naacl-main.272

    2021 doi

  9. [17]

    Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models. InThirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Trac...

    2021

  10. [18]

    Nandan Thakur, Nils Reimers, Andreas Ruckl’e, Abhishek Srivastava, and Iryna Gurevych. 2021. BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models

    2021

  11. [19]

    Robert W. White. 2016.Interactions with Search Systems. Cambridge University Press

    2016

  12. [20]

    Lijun Yao, Q. Sun, C. Wang, and Z. Ding. 2019. An Empirical Study on Cross- Domain Label Noise. InProceedings of the International Conference on Machine Learning. 7068–7077

    2019

  13. [21]

    Zhang, and Shaoping Ma

    Jingtao Zhan, Qingyao Ai, Yiqun Liu, Jiaxin Mao, Xiaohui Xie, M. Zhang, and Shaoping Ma. 2022. Disentangled Modeling of Domain and Relevance for Adapt- able Dense Retrieval. doi:10.48550/arXiv.2208.05753

    2022 doi arXiv

  14. [22]

    Wei Zhong, Yuqing Xie, and Jimmy J. Lin. 2022. Applying Structural and Dense Semantic Matching for the ARQMath Lab 2022, CLEF. InConference and Labs of the Evaluation Forum. https://api.semanticscholar.org/CorpusID:251471859 A Preliminary Experiments A.1 Additional Results for...

    2022

  15. [446]

    com/&output=citation&hl=de&as_sdt=0,5&ct=citation&cd=0

    http://scholar.google.de/scholar.bib?q=info:6Bdw8cs-UYMJ:scholar.google. com/&output=citation&hl=de&as_sdt=0,5&ct=citation&cd=0

  16. [2020]

    LEGAL-BERT: The Muppets Straight Out of Law School

  17. [2024]

    In2024 Medical Technologies Congress (TIPTEKNO)

    Analysis of Retrieval Performance for Methods Fine-Tuned with ColBERT Architecture. In2024 Medical Technologies Congress (TIPTEKNO). 1–4. doi:10. 1109/TIPTEKNO63488.2024.10755364

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.