Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Deep Retrieval at CheckThat! 2025: Identifying Scientific Papers from Implicit Social Media Mentions via Hybrid Retrieval and Re-Ranking

T0 review · 4 major / 4 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 A three-stage retrieval pipeline links social media posts to the scientific papers they implicitly reference.

desk verdict A solid shared-task system paper with credible external results; the missing recall ceiling and code release keep it from being more than conditionally useful. read the letter →

arxiv 2505.23250 v2 pith:3NOJN7HR submitted 2025-05-29 cs.IR cs.AI

Pascal J. Sager , Ashwini Kamaraj , Benjamin F. Grewe , Thilo Stadelmann This is my paper · ORCID
classification cs.IRcs.AI MSC 68P2068T50
keywords hybridretrievalscientificpapersocialmediafact-checkingdenseBM25cross-encoderre-rankingMRR@5CLEFCheckThat!2025
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 argues that no single retrieval method can bridge the gap between informal social media language and formal scientific abstracts, and that a hybrid pipeline can. It proposes combining BM25 lexical search, dense semantic search with a fine-tuned embedding model, and LLM-based cross-encoder re-ranking. On the shared task's development and test sets, this system achieves MRR@5 of 76.46% on the development set and 66.43% on the test set, ranking 1st on the dev leaderboard and 3rd on the test leaderboard out of 31 teams. The result matters because it shows competitive scientific source attribution can be done with open-source models, locally, and without external training data.

What carries the argument

The load-bearing mechanism is the three-stage hybrid pipeline. Documents are indexed as title-plus-abstract strings in two parallel retrievers: a BM25 sparse retriever with lowercasing, punctuation removal, and byte-pair subword tokenization returns the top 30 candidates; a FAISS vector store over embeddings from the INF-Retriever-v1 dense model, fine-tuned on the task's training set with multiple-negatives ranking loss, returns the top 100. The 130 candidates are deduplicated and scored jointly by bge-reranker-v2-gemma, an LLM-based cross-encoder, which outputs the final top five.

What would settle it

Take the official development or test queries, merge the top 30 BM25 candidates and the top 100 dense candidates for each query, and measure recall of the known correct paper within that 130-document set. A recall noticeably below 100 percent would show that candidate generation, not re-ranking, sets the upper bound on the reported MRR@5, and increasing or improving the candidate set would test whether re-ranking gains grow further.

Watch

Extended reading notes

Core claim

The central claim is that merging lexical and dense candidate sets and then re-ranking with a large-language-model cross-encoder substantially improves the ability to identify the correct scientific paper for a social media post. With BM25 alone the test MRR@5 is 51.47%; a fine-tuned semantic retriever reaches 56.72%; adding the cross-encoder re-ranker raises the full pipeline to 66.43% on the test set and 76.46% on the development set. The paper also reports that in-domain fine-tuning of the dense retriever adds about two percentage points and that reciprocal rank fusion, a lighter fusion baseline, underperforms the cross-encoder by about seven points.

Load-bearing premise

The correct paper must already be contained in the union of the top 30 BM25 hits and the top 100 dense hits for a query; if it is absent, the re-ranker cannot recover it, and the paper does not report how often this union contains the answer.

Editorial extensions

If this is right

  • Using no external training data, metadata, or commercial models, the pipeline reaches within two percentage points of the top test system, so strong source-attribution performance is attainable with open components.
  • Fine-tuning the dense retriever on in-domain claim-paper pairs yields consistent absolute gains of about two MRR@5 points across dev and test.
  • Cross-encoder re-ranking adds roughly nine to ten MRR@5 points over the best individual retriever, while reciprocal rank fusion adds only about two, indicating that learned joint scoring matters more than rank fusion.
  • LLM-based query expansion and rewriting improve lexical precision but were excluded because their 60x inference cost is not justified once re-ranking is in place.

Reading between the lines

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

  • Because the paper does not report recall of the merged 130-candidate set, the re-ranking gains it measures are a lower bound; a wider or better candidate set could push MRR@5 higher.
  • The same three-stage design should transfer to other informal-to-formal retrieval problems, such as linking patient forum posts to clinical studies or news headlines to policy documents, since the components are not task-specific.
  • Adding metadata such as author names, venues, and timestamps, which the paper explicitly leaves out, is a cheap extension that could improve disambiguation when several papers are near ties.
  • The augmentation experiments suggest that after fine-tuning, synthetic query and document expansion adds little; future gains may come from harder negatives or training the re-ranker itself rather than from more text variants.
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 / 4 minor

Summary. The paper presents the Deep Retrieval team's system for CLEF CheckThat! 2025 subtask 4b, which retrieves scientific papers relevant to social media posts. The pipeline combines a BM25 lexical retriever (top-30 candidates), a dense retriever using a fine-tuned INF-Retriever-v1 model (top-100 candidates), and an LLM-based cross-encoder re-ranker (bge-reranker-v2-gemma) over the merged 130-candidate set. The system achieves MRR@5 of 76.46% on the development set (1st place) and 66.43% on the hidden test set (3rd place out of 31 teams), without using additional task-specific labeled data beyond the official training set. The paper reports ablations on preprocessing, fine-tuning, re-ranking model choice, and data augmentation, and includes appendices with implementation details.

Significance. If the results hold, the paper provides a useful, fully open-source system description for a challenging retrieval task: matching informal social media language to formal scientific abstracts. The evaluation is on an external, fixed test set with a public leaderboard, which lends credibility to the main empirical claims. The paper includes reproducible engineering details, including exact hyperparameters for LoRA fine-tuning and prompt templates for augmentation experiments. The main contributions are the demonstrated consistent gains from lexical preprocessing, embedding fine-tuning, and LLM-based re-ranking across dev and test splits. The claims are, however, partially limited by missing candidate-recall statistics, single-run point estimates, and an overbroad 'no external training data' statement.

major comments (4)
  1. [§3.3, Table 1] The recall of the merged candidate set (union of BM25 top-30 and dense top-100) is never reported, so the upper bound on the re-ranking stage is not quantified. Since the re-ranker only reorders the retrieved candidates, the final MRR@5 cannot exceed Recall@130 of this merged set. Precision@30 and Precision@100 of the individual retrievers do not bound this recall unless each query has exactly one relevant document, a condition the paper never states (the task description and Section 4 vacillate between 'correct source' singular and 'scientific papers' plural). Please report Recall@130 on dev and test, or explicitly justify the single-relevant-per-query assumption. Without this, the +9.3–9.7 MRR@5 gain attributed to re-ranking is an observed lower bound whose headroom is unknown, and the claim that 'hybrid retrieval is essential' is not fully supported.
  2. [§4, Table 3] All experimental results are single-run point estimates with no variance, confidence intervals, or significance tests. This is particularly consequential in Appendix C, where bge-reranker-v2-gemma is selected over bge-reranker-v2-minicpm based on a 0.01 point MRR@5 difference (76.03% vs 76.02%). Such a difference is within noise. If the final model choice in Section 3.3 rests on this comparison, the authors should provide multiple runs with standard deviations (at least on the development set) or explicitly state that the difference is negligible and selection was based on other criteria.
  3. [Abstract; §1] The claim of achieving strong performance 'without external training data' is ambiguous and, taken literally, false: the pipeline builds on pre-trained models (INF-Retriever-v1, BAAI/bge-reranker-v2-gemma) whose parameters were trained on external corpora. What the authors mean is that no additional task-specific labeled data beyond the official CheckThat! training set was used. This distinction should be stated explicitly in the Abstract and throughout, because the 'no external data' claim is a headline contribution and currently overstates the resource independence of the approach.
  4. [§4, Appendix E] The comparison with the Elasticsearch RRF baseline is not apples-to-apples: the Elasticsearch pipeline uses different retrievers (BM25 with fuzzy matching, kNN with k=50/200), a different number of candidates, and is only evaluated on the development set. The 7.1-point gap between RRF and the cross-encoder therefore conflates the re-ranking method with differences in candidate generation and fusion configuration. The direct comparison of the full pipeline against the best individual retriever (same candidate set) is more convincing; the RRF comparison should either be aligned to the same candidate set or be reported as a separate system variant rather than as evidence for the added value of the cross-encoder.
minor comments (4)
  1. [Title page/footnote] The corresponding author email in the footnote appears garbled as '/envel⌢pe-⌢penpascaljosef.sager@uzh.ch'; this should be corrected to a proper email address.
  2. [§3.3, Appendix C] Table 3 in Appendix C evaluates re-rankers on top-50 semantic retrieval candidates, whereas the final pipeline re-ranks the union of top-100 dense and top-30 BM25 candidates. Please clarify why the re-ranker selection was performed on a different candidate set and whether the ordering of models is stable on the final candidate set.
  3. [Throughout] The embedding model name is written inconsistently as 'INF-Retriever-v1', 'Inf-Retriever-v1', and 'inf-retriever-v1' (e.g., §3.2 vs. reference [10]); please unify the spelling.
  4. [Table 1] Table 1 shows the Elasticsearch RRF row only for the development set, but the table caption does not state this. Please add a note that Elasticsearch was not evaluated on the test set.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central results are externally evaluated on official dev/test splits with fixed ground truth, and the only self-citation is not load-bearing.

full rationale

The paper's central claim is an empirical performance result on the official CheckThat! 2025 Subtask 4b benchmark, measured by MRR@5 against a fixed external ground truth. The pipeline derivation—BM25 lexical retrieval, FAISS-based dense retrieval with INF-Retriever-v1, and bge-reranker-v2-gemma cross-encoder re-ranking—does not define any target quantity in terms of the reported metric. The dense retriever is fine-tuned on the official training set using multiple negatives ranking loss, and model choices and hyperparameters are selected using the development set; the final test-set numbers are independent of that selection. The only author self-citation is reference [29], used to support a general statement that LoRA reduces memory and training time; this claim is not load-bearing for the retrieval results. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. The absence of a reported Recall@130 for the merged 130-candidate set is a completeness and ceiling-analysis gap, but it is not circularity: the pipeline's components are evaluated against an external benchmark rather than against their own outputs. Accordingly, the appropriate finding is no significant circularity.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

This is an empirical systems paper, so the ledger lists the hand-tuned hyperparameters, the dataset and metric assumptions, and the base-model assumptions. No new theoretical entities are introduced.

free parameters (9)
  • BM25 candidate count (top_k=30) = 30
    Chosen by hand; the paper states increasing BM25 candidates beyond 30 did not improve re-ranking but increased cost (Sec. 3.3).
  • Dense candidate count (top_k=100) = 100
    Selected empirically; increasing dense candidates to 100 led to substantial gains (Sec. 3.3).
  • LoRA rank = 8
    LoRA hyperparameter for fine-tuning the INF-Retriever-v1 embedding model (Appendix B).
  • LoRA alpha = 32
    LoRA scaling hyperparameter (Appendix B).
  • LoRA dropout = 0.1
    LoRA dropout ratio (Appendix B).
  • Learning rate = 1e-5
    AdamW learning rate for fine-tuning (Appendix B).
  • Effective batch size = 64
    Batch size for MNR fine-tuning (Appendix B).
  • Number of epochs = 2
    Checkpoint selected by dev Precision@100, best at epoch 2 (Appendix B).
  • Re-ranker model choice = bge-reranker-v2-gemma
    Selected based on dev MRR@5 among four re-rankers (Table 3); margin over alternatives is 0.01-0.16 points.
assumptions (5)
  • domain assumption The official CheckThat! 2025 subtask 4b ground truth labels are correct and complete.
    All evaluation numbers are computed against this corpus (Sec. 4).
  • domain assumption MRR@5 is the agreed evaluation metric for the task.
    The competition defines the leaderboard by MRR@5; the paper adopts it without justification.
  • domain assumption The pretrained models (INF-Retriever-v1, bge-reranker-v2-gemma) are suitable base models for cross-domain retrieval and re-ranking.
    The paper chooses these models based on prior performance and open availability; no proof that they are optimal, but their behavior is treated as given (Secs. 3.2, 3.3).
  • domain assumption The fine-tuning procedure on the official training set with MNR loss improves retrieval on the test distribution.
    Empirically validated on dev/test (Table 1), but it is an assumption that the training set is representative and the objective aligns with MRR@5.
  • domain assumption Title and abstract alone are sufficient document representations for this retrieval task.
    The paper deliberately excludes metadata (authors, venue) and reports that metadata could improve results (Sec. 3, Sec. 5 limitations).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Retrieval at CheckThat! 2025: Identifying Scientific Papers from Implicit Social Media Mentions via Hybrid Retrieval and Re-Ranking." pith.science (2026). https://pith.science/paper/3NOJN7HR

@misc{pith2026250523250,
  author       = {Pith},
  title        = {Pith review of: Deep Retrieval at CheckThat! 2025: Identifying Scientific Papers from Implicit Social Media Mentions via Hybrid Retrieval and Re-Ranking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3NOJN7HR}},
  note         = {Machine review of arXiv:2505.23250}
}
read the original abstract

We present the methodology and results of the Deep Retrieval team for subtask 4b of the CLEF CheckThat! 2025 competition, which focuses on retrieving relevant scientific literature for given social media posts. To address this task, we propose a hybrid retrieval pipeline that combines lexical precision, semantic generalization, and deep contextual re-ranking, enabling robust retrieval that bridges the informal-to-formal language gap. Specifically, we combine BM25-based keyword matching with a FAISS vector store using a fine-tuned INF-Retriever-v1 model for dense semantic retrieval. BM25 returns the top 30 candidates, and semantic search yields 100 candidates, which are then merged and re-ranked via a large language model (LLM)-based cross-encoder. Our approach achieves a mean reciprocal rank at 5 (MRR@5) of 76.46% on the development set and 66.43% on the hidden test set, securing the 1st position on the development leaderboard and ranking 3rd on the test leaderboard (out of 31 teams), with a relative performance gap of only 2 percentage points compared to the top-ranked system. We achieve this strong performance by running open-source models locally and without external training data, highlighting the effectiveness of a carefully designed and fine-tuned retrieval pipeline.

Figures

Figures reproduced from arXiv: 2505.23250 by the authors.

Figure 1
Figure 1. The top of the figure displays the task of finding relevant publications from a large pool of scientific documents, given a social media post. The bottom part of the figure provides an overview of our pipeline, consisting of three modules to predict the top five scientific documents ordered by their importance. 3. Re-ranking with a large language model (LLM)-based cross-encoder [11, 12], which jointly encodes and sc… view at source ↗
Figure 2
Figure 2. Overview of our retrieval pipeline. Scientific documents are indexed using two parallel retrieval mechanisms: A lexical retriever and a semantic retriever. Given a query (e.g., a social media post), the lexical retriever returns 30 candidate papers, and the semantic retriever outputs its top 100 candidates. These 130 candidates are then re-ranked using a cross-encoder. from social media, due to the frequent lexical … view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Prompting Beats Fine-Tuning: Generative Expected Value Scoring for Statutory Term Retrieval

    cs.IR 2026-07 conditional novelty 4.0 of 10

    Generative expected-value prompting of decoder-only LLMs beats fine-tuned ModernBERT and prior SOTA on ranking 26,959 case-law sentences for 42 U.S. Code terms.

  2. SciClaimSeekers at CheckThat! 2026: Retrieving Scientific Sources for Social Media Claims with LLM Reranking

    cs.IR 2026-07 conditional novelty 4.0 of 10

    A zero-shot Qwen2.5-14B reranker on top of BM25+E5 hybrid retrieval reaches 64.39 MRR@5 on CLEF-2026 CheckThat! Task 1 English scientific source retrieval, with the LLM contributing most of the gain.

Reference graph

Works this paper leans on

40 extracted references · 15 canonical work pages · cited by 2 Pith papers

  1. [1]

    Brüggemann, I

    M. Brüggemann, I. Lörcher, S. Walter, Post-normal science communication: Exploring the blurring boundaries of science and journalism, Journal of Science Communication 19 (2020) A02. doi:10. 22323/2.19030202

    work page 2020

  2. [2]

    F. Alam, S. Shaar, F. Dalvi, H. Sajjad, A. Nikolov, H. Mubarak, et al., Fighting the COVID-19 Infodemic: Modeling the Perspective of Journalists, Fact-Checkers, Social Media Platforms, Policy Makers, and the Society, in: Findings of the Association for Computational Linguistics: EMNLP 2021, Association for Computational Linguistics, Punta Cana, Dominican ...

    doi:10.18653/v1/2021.findings-emnlp.56 2021

  3. [3]

    F. Alam, J. M. Struß, T. Chakraborty, S. Dietze, S. Hafid, K. Korre, A. Muti, P. Nakov, F. Ruggeri, S. Schellhammer, V. Setty, M. Sundriyal, K. Todorov, V. V., The CLEF-2025 CheckThat! Lab: Subjectivity, Fact-Checking, Claim Normalization, and Retrieval, in: C. Hauff, C. Macdonald, D. Jannach, G. Kazai, F. M. Nardini, F. Pinelli, F. Silvestri, N. Tonellot...

    work page 2025

  4. [4]

    F. Alam, J. M. Struß, T. Chakraborty, S. Dietze, S. Hafid, K. Korre, A. Muti, P. Nakov, F. Ruggeri, S. Schellhammer, V. Setty, M. Sundriyal, K. Todorov, V. Venktesh, Overview of the CLEF-2025 CheckThat! Lab: Subjectivity, Fact-Checking, Claim Normalization, and Retrieval, in: J. Carrillo-de Albornoz, J. Gonzalo, L. Plaza, A. García Seco de Herrera, J. Mot...

    work page 2025

  5. [5]

    Hafid, Y

    S. Hafid, Y. S. Kartal, S. Schellhammer, K. Boland, D. Dimitrov, S. Bringay, K. Todorov, S. Dietze, Overview of the CLEF-2025 CheckThat! Lab Task 4 on Scientific Web Discourse, in: G. Faggioli, N. Ferro, P. Rosso, D. Spina (Eds.), Working Notes of CLEF 2025 - Conference and Labs of the Evaluation Forum, CLEF 2025, Madrid, Spain, 2025

    work page 2025

  6. [6]

    S. E. Robertson, K. S. Jones, Relevance weighting of search terms, Journal of the American Society for Information Science 27 (1976) 129–146. doi:10.1002/asi.4630270302

    doi:10.1002/asi.4630270302 1976

  7. [7]

    Robertson, H

    S. Robertson, H. Zaragoza, The Probabilistic Relevance Framework: BM25 and Beyond, Foundations and Trends® in Information Retrieval 3 (2009) 333–389. doi:10.1561/1500000019

    doi:10.1561/1500000019 2009

  8. [8]

    Johnson, M

    J. Johnson, M. Douze, H. Jegou, Billion-Scale Similarity Search with GPUs, IEEE Transactions on Big Data 7 (2021) 535–547. doi:10.1109/TBDATA.2019.2921572. 2https://chatgpt.com 3https://www.grammarly.com

    arXiv 2021

Show all 40 references
  1. [9]

    Douze, A

    M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.-E. Mazaré, et al., The Faiss library, 2024. doi:10.48550/ARXIV.2401.08281, version Number: 3

    2024 doi arXiv

  2. [10]

    J. Yang, J. Wan, Y. Yao, W. Chu, Y. Xu, et al., inf-retriever-v1 (2025). URL: https://huggingface.co/ infly/inf-retriever-v1. doi:10.57967/HF/4262

    2025 doi

  3. [11]

    C. Li, Z. Liu, S. Xiao, Y. Shao, Making Large Language Models A Better Foundation For Dense Retrieval, 2023. doi:10.48550/ARXIV.2312.15503, version Number: 1

    2023 doi

  4. [12]

    J. Chen, S. Xiao, P. Zhang, K. Luo, D. Lian, Z. Liu, M3-Embedding: Multi-Linguality, Multi- Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation, in: Findings of the Association for Computational Linguistics ACL 2024, Association for Computation...

    2024 doi

  5. [13]

    Thorne, A

    J. Thorne, A. Vlachos, C. Christodoulopoulos, A. Mittal, FEVER: a Large-scale Dataset for Fact Extraction and VERification, in: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volum...

    2018 doi

  6. [14]

    Vlachos, S

    A. Vlachos, S. Riedel, Fact Checking: Task definition and dataset construction, in: Proceedings of the ACL 2014 Workshop on Language Technologies and Computational Social Science, Association for Computational Linguistics, Baltimore, MD, USA, 2014, pp. 18–22. doi:10.3115/v1/W14-2508

    2014 doi

  7. [15]

    Wadden, S

    D. Wadden, S. Lin, K. Lo, L. L. Wang, M. Van Zuylen, A. Cohan, et al., Fact or Fiction: Verifying Scientific Claims, in: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), Association for Computational Linguistics, Online, 2020, pp....

    2020 doi

  8. [16]

    Wadden, K

    D. Wadden, K. Lo, B. Kuehl, A. Cohan, I. Beltagy, L. L. Wang, et al., SciFact-Open: Towards open-domain scientific claim verification, in: Findings of the Association for Computational Linguistics: EMNLP 2022, Association for Computational Linguistics, Abu Dhabi, United Arab E...

    2022 doi

  9. [17]

    Barrón-Cedeño, F

    A. Barrón-Cedeño, F. Alam, A. Galassi, G. Da San Martino, P. Nakov, T. Elsayed, et al., Overview of the CLEF–2023 CheckThat! Lab on Checkworthiness, Subjectivity, Political Bias, Factuality, and Authority of News Articles and Their Source, in: Experimental IR Meets Multilingua...

    2023

  10. [18]

    Barrón-Cedeño, F

    A. Barrón-Cedeño, F. Alam, J. M. Struß, P. Nakov, T. Chakraborty, T. Elsayed, et al., Overview of the CLEF-2024 CheckThat! Lab: Check-Worthiness, Subjectivity, Persuasion, Roles, Authorities, and Adversarial Robustness, in: Experimental IR Meets Multilinguality, Multimodality,...

    2024

  11. [19]

    Karpukhin, B

    V. Karpukhin, B. Oguz, S. Min, P. Lewis, L. Wu, S. Edunov, et al., Dense Passage Retrieval for Open-Domain Question Answering, in: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), Association for Computational Linguistics, Online,...

    2020 doi

  12. [20]

    Izacard, E

    G. Izacard, E. Grave, Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering, in: Proceedings of the 16th Conference of the European Chapter of the As- sociation for Computational Linguistics: Main Volume, Association for Computational Linguisti...

    2021 doi

  13. [21]

    J. Lee, M. Sung, J. Kang, D. Chen, Learning Dense Representations of Phrases at Scale, in: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)...

    2021 doi

  14. [22]

    Maillard, V

    J. Maillard, V. Karpukhin, F. Petroni, W.-t. Yih, B. Oguz, V. Stoyanov, et al., Multi-Task Retrieval for Knowledge-Intensive Tasks, in: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natura...

    2021 doi

  15. [23]

    Nogueira, K

    R. Nogueira, K. Cho, Passage Re-ranking with BERT, 2019. doi:10.48550/ARXIV.1901.04085, version Number: 5

    2019 doi arXiv

  16. [24]

    Sennrich, B

    R. Sennrich, B. Haddow, A. Birch, Neural Machine Translation of Rare Words with Subword Units, in: Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Association for Computational Linguistics, Berlin, Germany, 2016,...

    2016 doi

  17. [25]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, et al., Atten- tion is All you Need, in: Advances in Neural Information Processing Systems, volume 30, Curran Associates, Inc., 2017. URL: https://proceedings.neurips.cc/paper_files/paper/2017/file/ 3f5ee2...

    2017

  18. [26]

    A. Yang, B. Yang, B. Hui, B. Zheng, B. Yu, C. Zhou, et al., Qwen2 Technical Report, 2024. doi:10. 48550/arXiv.2407.10671, arXiv:2407.10671 [cs]

    2024 doi arXiv

  19. [27]

    Henderson, R

    M. Henderson, R. Al-Rfou, B. Strope, Y.-h. Sung, L. Lukacs, R. Guo, et al., Efficient Natu- ral Language Response Suggestion for Smart Reply, 2017. doi:10.48550/arXiv.1705.00652, arXiv:1705.00652 [cs]

    2017 doi arXiv

  20. [28]

    E. J. Hu, y. shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, et al., LoRA: Low-Rank Adaptation of Large Language Models, in: International Conference on Learning Representations, 2022. URL: https://openreview.net/forum?id=nZeVKeeFYf9

    2022

  21. [29]

    Tuggener, P

    L. Tuggener, P. Sager, Y. Taoudi-Benchekroun, B. F. Grewe, T. Stadelmann, So you want your private LLM at home? A survey and benchmark of methods for efficient GPTs, in: 2024 11th IEEE Swiss Conference on Data Science (SDS), IEEE, Zurich, Switzerland, 2024, pp. 205–212. doi:10...

    2024

  22. [30]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Decoupled Weight Decay Regularization, in: International Conference on Learning Representations, 2019. URL: https://openreview.net/forum?id=Bkg6RiCqY7

    2019

  23. [31]

    Riviere, S

    Gemma Team, M. Riviere, S. Pathak, P. G. Sessa, C. Hardin, et al., Gemma 2: Improving Open Language Models at a Practical Size, 2024. URL: https://arxiv.org/abs/2408.00118. doi:10.48550/ ARXIV.2408.00118, version Number: 3

    2024 doi arXiv

  24. [32]

    G. V. Cormack, C. L. A. Clarke, S. Buettcher, Reciprocal rank fusion outperforms condorcet and individual rank learning methods, in: Proceedings of the 32nd international ACM SIGIR conference on Research and development in information retrieval, ACM, Boston MA USA, 2009, pp. 7...

    2009

  25. [33]

    Kamath, J

    Gemma Team, A. Kamath, J. Ferret, S. Pathak, N. Vieillard, R. Merhej, S. Perrin, T. Matejovicova, A. Ramé, M. Rivière, L. Rouillard, T. Mesnard, G. Cideron, J. bastien Grill, S. Ramos, E. Yvinec, M. Casbon, E. Pot, I. Penchev, G. Liu, F. Visin, K. Kenealy, L. Beyer, X. Zhai, A...

    2025 doi arXiv

  26. [34]

    Shakir, D

    A. Shakir, D. Koenig, J. Lipp, S. Lee, Boost Your Search With The Crispy Mixedbread Rerank Models, 2024. URL: https://www.mixedbread.ai/blog/mxbai-rerank-v1

    2024

  27. [35]

    S. Xiao, Z. Liu, P. Zhang, N. Muennighoff, C-pack: Packaged resources to advance general chinese embedding, 2023. arXiv:2309.07597

    2023 arXiv

  28. [36]

    L. Gao, X. Ma, J. Lin, J. Callan, Precise Zero-Shot Dense Retrieval without Relevance Labels, in: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Association for Computational Linguistics, Toronto, Canada, 2023, ...

    2023 doi

  29. [37]

    Grattafiori, A

    A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, et al., The Llama 3 Herd of Models, 2024. doi:10.48550/ARXIV.2407.21783, version Number: 3. A. Additional Experiments on Lexical Retrieval Table 2 Comparison of lexical retrieval approaches with optional q...

    2024 doi arXiv

  30. [39]

    Query Rewriting: Reformulating social media posts to correct grammar and match the formal language style of scientific abstracts while preserving the original query semantics (see Listing 1)

  31. [40]

    Just saw amazin new study - mice w/ #Alzheimers showed 45% improvemnt in memory after new drug treatment!! Game changer for #neurodegeneration research imo

    Query Expansion: Augmenting the original social media post with 2-3 contextually relevant sentences to increase n-gram overlap with scientific abstracts (see Listing 2). Among the evaluated methods, query expansion yielded the highest performance, achieving a Preci- sion@20 of...

  32. [275]

    doi:10.1007/978-3-031-42448-9_20 , series Title: Lecture Notes in Computer Science

    doi

Pith tools

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