pith. sign in

arxiv: 2605.28522 · v1 · pith:4XTGDXEKnew · submitted 2026-05-27 · 💻 cs.IR

Search for Coverage: Learning Coverage-Aware Retrieval with Augmented Sub-Question Answerability

Jia-Huei Ju , Eugene Yang , Trevor Adriaanse , Suzan Verberne , Andrew Yates This is my paper

Pith reviewed 2026-06-29 09:45 UTC · model grok-4.3

classification 💻 cs.IR
keywords coverage-aware retrievaldense retrievalretrieval-augmented generationcontrastive learningsub-question answerabilitylong-form QAnugget coverage
0
0 comments X

The pith

CoveR is a bi-encoder trained on synthetic coverage signals from LLM sub-question judgments that raises nugget coverage by 10% over dense baselines.

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

The paper introduces CoveR, a dense retrieval model built for coverage-aware ranking in long-form RAG. CoveR uses coverage-based contrastive and distillation objectives on the SCOPE dataset of 90K pairs, where coverage signals come from LLM-generated sub-question answerability judgments added to Researchy Questions. The goal is to make the retriever capture diverse aspects of a query while preserving standard relevance performance. A sympathetic reader would care because long-form generation needs all relevant facts retrieved, not just the most similar ones, and the reported 10% coverage gain occurs without relevance loss.

Core claim

CoveR is a bi-encoder trained with the coverage-based contrastive and distillation objectives on the SCOPE dataset, which comprises 90K training pairs from Researchy Questions with synthetic coverage signals augmented from sub-question answerability judgments generated by LLMs. This training enables CoveR to capture diverse aspects of information needs, and empirical experiments show that CoveR enhances nugget coverage by 10% over strong dense retrieval baselines without sacrificing its relevance-based retrieval capability.

What carries the argument

Coverage-based contrastive and distillation objectives in a bi-encoder that use synthetic coverage signals from LLM sub-question answerability judgments.

If this is right

  • Nugget coverage rises by 10% relative to strong dense retrieval baselines.
  • Relevance-based ranking performance stays intact.
  • The model produces a better relevance-coverage trade-off than prior dense retrievers.
  • Ablation studies confirm that both the contrastive and distillation objectives contribute to the observed gains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same synthetic labeling approach could be reused for other retrieval settings where explicit coverage annotations are expensive to collect.
  • Downstream RAG pipelines that rely on this retriever may produce fewer incomplete answers because more nuggets reach the generator.
  • The method could be tested by swapping the LLM labeler for a smaller model or for human sub-question judgments to measure sensitivity to label quality.

Load-bearing premise

LLM-generated sub-question answerability judgments supply reliable synthetic coverage signals that match human notions of comprehensive retrieval.

What would settle it

A side-by-side human evaluation of retrieved sets in which the correlation between the LLM-derived coverage labels and human judgments of fact completeness falls below a statistically significant threshold.

Figures

Figures reproduced from arXiv: 2605.28522 by Andrew Yates, Eugene Yang, Jia-Huei Ju, Suzan Verberne, Trevor Adriaanse.

Figure 1
Figure 1. Figure 1: Effectiveness for relevance (𝑥-axis) vs. information coverage (𝑦-axis). Relevance is reported with the average nDCG@10 across 13 BEIR datasets. Information coverage is reported with the average 𝛼-nDCG@10 across 3 nugget-based retrieval evaluation datasets. 1 Introduction As LLMs have evolved to support longer contexts, a new search par￾adigm has emerged. Beyond presenting search results via a simple rankin… view at source ↗
Figure 2
Figure 2. Figure 2: The two proposed coverage-based training methods: coverage contrastive (CovCon) and coverage self-distillation [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The instruction prompt used for generating distil [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The accumulated Coverage scores of the top- [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: Rubric-based answerability judgment prompt. The [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: The performance changes over Unsupervised base￾line with different sampling ranges for positive and negatives. The criteria of two ranges are based on the estimated cover￾age score of documents. The upper parenthesis (𝛼1, 𝛼2) is the range for high coverage documents; the lower parenthesis (𝛽1, 𝛽2.) is the range for low coverage documents [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The prompt used for generating sub-questions. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
read the original abstract

Long-form Retrieval-Augmented Generation (RAG) brings the challenge of coverage-based ranking, because ranking methods must ensure the inclusion of comprehensive relevant nuggets (i.e., facts), which can thereby be synthesized into a comprehensive output. In this work, we propose CoveR (Our code is available at https://github.com/DylanJoo/CoveR ) a dense retrieval method optimized for coverage-aware retrieval scenarios. CoveR is a bi-encoder trained with the coverage-based contrastive and distillation objectives, which enables CoveR to capture diverse aspects of information needs. To train CoveR, we create the SCOPE dataset, (Our training data is available at https://huggingface.co/datasets/DylanJHJ/scope ) which comprises 90K training pairs from Researchy Questions with synthetic coverage signals augmented from sub-question answerability judgments generated by LLMs. Our empirical experiments show that CoveR enhances nugget coverage by 10\% over strong dense retrieval baselines without sacrificing its relevance-based retrieval capability. Further ablation studies validate the importance of our proposed learning method, showing that CoveR achieves a superior trade-off between relevance- and coverage-based ranking, which is essential for long-form RAG.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes CoveR, a bi-encoder dense retriever trained via coverage-aware contrastive and distillation objectives to improve nugget coverage for long-form RAG. It introduces the SCOPE dataset (90K pairs derived from Researchy Questions) whose coverage signals are generated synthetically via LLM judgments of sub-question answerability. The central empirical claim is a 10% lift in nugget coverage over strong dense baselines while preserving relevance-based ranking performance; code and training data are released.

Significance. If the LLM-derived coverage signals prove to be reliable proxies, the work supplies a practical training recipe for coverage-aware retrieval, an under-addressed requirement in comprehensive RAG pipelines. The public release of both the model code and the SCOPE dataset constitutes a concrete reproducibility asset that strengthens the contribution.

major comments (2)
  1. [Abstract] Abstract (dataset construction paragraph): the 10% coverage gain is obtained on labels produced solely by LLM sub-question answerability judgments. No human validation, inter-annotator agreement, or correlation study with human notions of nugget coverage is reported; because the central claim rests on these synthetic signals being faithful proxies, the absence of such validation is load-bearing.
  2. [Empirical experiments] Empirical experiments (results section): the abstract states a 10% coverage improvement but supplies neither error bars, statistical significance tests, nor visible ablation tables or dataset statistics. Without these, the claim that CoveR achieves a superior relevance-coverage trade-off cannot be verified from the reported evidence.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the validation of synthetic signals and the reporting of experimental results. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (dataset construction paragraph): the 10% coverage gain is obtained on labels produced solely by LLM sub-question answerability judgments. No human validation, inter-annotator agreement, or correlation study with human notions of nugget coverage is reported; because the central claim rests on these synthetic signals being faithful proxies, the absence of such validation is load-bearing.

    Authors: We agree that the absence of human validation for the LLM-generated coverage signals is a limitation, as these signals underpin the reported gains. Our experiments demonstrate consistent improvements when training and evaluating against the same synthetic labels, and we release the full SCOPE dataset to support future human correlation studies. We will add an explicit discussion of this limitation in the revised Limitations section. revision: partial

  2. Referee: [Empirical experiments] Empirical experiments (results section): the abstract states a 10% coverage improvement but supplies neither error bars, statistical significance tests, nor visible ablation tables or dataset statistics. Without these, the claim that CoveR achieves a superior relevance-coverage trade-off cannot be verified from the reported evidence.

    Authors: We acknowledge that the current reporting lacks error bars, significance tests, and expanded dataset statistics. The manuscript already contains ablation studies on the learning objectives, but we will revise the results section to include error bars on main tables, paired statistical significance tests, SCOPE dataset statistics (e.g., sub-question distributions), and clearer ablation tables to substantiate the relevance-coverage trade-off. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper constructs the SCOPE dataset from Researchy Questions using LLM-generated sub-question answerability judgments as synthetic coverage signals, then trains a bi-encoder via standard coverage-based contrastive and distillation objectives. The central empirical claim (10% nugget coverage gain over dense baselines) is presented as an experimental outcome rather than a mathematical identity or fitted parameter renamed as prediction. No equations, self-definitional steps, or load-bearing self-citations appear in the abstract or dataset description that would reduce the result to its inputs by construction. The training pipeline relies on external LLM signals and conventional losses, leaving the reported improvements independently falsifiable via held-out evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the validity of LLM sub-question judgments as coverage proxies and on the assumption that the chosen contrastive and distillation losses directly optimize the reported coverage metric; no free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption LLM-generated answerability labels for sub-questions serve as faithful proxies for human coverage requirements
    Invoked when constructing the SCOPE training pairs from Researchy Questions

pith-pipeline@v0.9.1-grok · 5753 in / 1230 out tokens · 28101 ms · 2026-06-29T09:45:32.688372+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 13 canonical work pages · 4 internal anchors

  1. [1]

    Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, et al

  2. [2]

    MS MARCO: A Human Generated MAchine Reading COmprehension Dataset

    MS MARCO: A human generated machine reading comprehension dataset. arXiv:1611.09268

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    Jaime Carbonell and Jade Goldstein. 1998. The use of MMR, diversity-based reranking for reordering documents and producing summaries. InProc. of SIGIR. 335–336

    1998

  4. [4]

    Hung-Ting Chen and Eunsol Choi. 2025. Open-World Evaluation for Retrieving Diverse Perspectives. InProc. of NAACL-HLT

    2025

  5. [5]

    Hung-Ting Chen, Xiang Liu, Shauli Ravfogel, and Eunsol Choi. 2025. Beyond Single Embeddings: Capturing Diverse Targets with Multi-Query Retrieval. arXiv:2511.02770

    work page arXiv 2025

  6. [6]

    Jianlyu Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu

  7. [7]

    InFindings of ACL

    M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation. InFindings of ACL. 2318– 2335

  8. [8]

    Charles L A Clarke, Maheedhar Kolla, Gordon V Cormack, Olga Vechtomova, Azin Ashkan, Stefan Büttcher, and Ian MacKinnon. 2008. Novelty and diversity in information retrieval evaluation. InProc. of SIGIR. 659–666

    2008

  9. [9]

    Cormack, Charles L A Clarke, and Stefan Buettcher

    Gordon V. Cormack, Charles L A Clarke, and Stefan Buettcher. 2009. Reciprocal rank fusion outperforms condorcet and individual rank learning methods. In Proc. of SIGIR. 758–759

    2009

  10. [10]

    Nick Craswell, Bhaskar Mitra, Emine Yilmaz, Daniel Campos, and Jimmy Lin

  11. [11]

    Craswell, B

    Overview of the TREC 2021 deep learning track. arXiv:2507.08191

    work page arXiv 2021

  12. [12]

    Lawrie Dawn, Macavaney Sean, Mayfield James, Mcnamee Paul, W Oard Douglas, Soldaini Luca, and Yang Eugene. 2025. Overview of the TREC 2024 NeuCLIR track. arXiv:2509.14355

    work page arXiv 2025

  13. [13]

    Laura Dietz. 2024. A workbench for autograding retrieve/generate systems. In Proc. of SIGIR

    2024

  14. [14]

    Alexander Fabbri, Irene Li, Tianwei She, Suyi Li, and Dragomir Radev. 2019. Multi- News: A Large-Scale Multi-Document Summarization Dataset and Abstractive Hierarchical Model. InProc. of ACL. 1074–1084

    2019

  15. [15]

    Naghmeh Farzi and Laura Dietz. 2024. An exam-based evaluation approach beyond traditional relevance judgments. arXiv:2402.00309

    work page arXiv 2024

  16. [16]

    Naghmeh Farzi and Laura Dietz. 2024. Pencils down! Automatic rubric-based evaluation of retrieve/generate systems. InProc. of SIGIR. 175–184

    2024

  17. [17]

    Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023. Enabling large language models to generate text with citations. InProc. of EMNLP. 6465–6488

    2023

  18. [18]

    Max Grusky, Mor Naaman, and Yoav Artzi. 2018. Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies. InProc. of NAACL-HLT. 708–719

    2018

  19. [19]

    Matthew Henderson, Rami Al-Rfou, Brian Strope, Yun hsuan Sung, Laszlo Lukacs, Ruiqi Guo, Sanjiv Kumar, Balint Miklos, and Ray Kurzweil. 2017. Efficient Natural Language Response Suggestion for Smart Reply. arXiv:1705.00652

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  20. [20]

    Sebastian Hofstätter, Sophia Althammer, Michael Schröder, Mete Sertkan, and Allan Hanbury. 2021. Improving Efficient Neural Ranking Models with Cross- Architecture Knowledge Distillation. arXiv:2010.02666

    work page arXiv 2021

  21. [21]

    Landry, Eugene Yang, Suzan Verberne, and Andrew Yates

    Jia-Huei Ju, François G. Landry, Eugene Yang, Suzan Verberne, and Andrew Yates

  22. [22]

    LANCER: LLM Reranking for Nugget Coverage. InProc. of ECIR. 188–203

  23. [23]

    Jia-Huei Ju, Suzan Verberne, Maarten de Rijke, and Andrew Yates. 2025. Con- trolled retrieval-augmented context evaluation for long-form RAG. InFindings of EMNLP. 21102–21121

    2025

  24. [24]

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense Passage Retrieval for Open- Domain Question Answering. InProc. of EMNLP. 6769–6781

    2020

  25. [25]

    Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. Natural Questions: A Benchmark for Question Answering Research.Tr...

    2019

  26. [26]

    Dahyun Lee, Yongrae Jo, Haeju Park, and Moontae Lee. 2025. Shifting from ranking to set selection for retrieval augmented generation. InProc. of ACL. 17606–17619. Search for Coverage: Learning Coverage-Aware Retrieval with Augmented Sub-Question Answerability SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia

    2025

  27. [27]

    Kenton Lee, Ming-Wei Chang, and Kristina Toutanova. 2019. Latent retrieval for weakly supervised open domain question answering. InProc. of ACL. 6086–6096

    2019

  28. [28]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks.Proc. of NIPS

    2020

  29. [29]

    Zhicong Li, Jiahao Wang, Zhishu Jiang, Hangyu Mao, Zhongxia Chen, Jiazhen Du, Yuanxing Zhang, Fuzheng Zhang, Di Zhang, and Yong Liu. 2024. DMQR-RAG: Diverse Multi-Query Rewriting for RAG. arXiv:arXiv:2411.13154

    work page arXiv 2024

  30. [30]

    Sheng-Chieh Lin, Akari Asai, Minghan Li, Barlas Oguz, Jimmy Lin, Yashar Mehdad, Wen tau Yih, and Xilun Chen. 2023. How to Train Your Dragon: Diverse Augmentation Towards Generalizable Dense Retrieval. InProc. of EMNLP

    2023

  31. [31]

    Xueguang Ma, Luyu Gao, Shengyao Zhuang, Jiaqi Samantha Zhan, Jamie Callan, and Jimmy Lin. 2025. Tevatron 2.0: Unified Document Retrieval Toolkit across Scale, Language, and Modality. InProc. of SIGIR. 4061–4065

    2025

  32. [32]

    James Mayfield, Eugene Yang, Dawn Lawrie, Sean MacAvaney, Paul McNamee, Douglas W Oard, Luca Soldaini, Ian Soboroff, Orion Weller, Efsun Kayi, Kate Sanders, Marc Mason, and Noah Hibbler. 2024. On the evaluation of machine- generated reports. InProc. of SIGIR. 1904–1915

    2024

  33. [33]

    MetaAI. 2024. The Llama 3 herd of models. arXiv:2407.21783

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  34. [34]

    Sewon Min, Kenton Lee, Ming-Wei Chang, Kristina Toutanova, and Hannaneh Hajishirzi. 2021. Joint passage ranking for diverse multi-answer retrieval. InProc. of EMNLP

    2021

  35. [35]

    Zach Nussbaum, John Xavier Morris, Andriy Mulyar, and Brandon Duderstadt

  36. [36]

    Nomic Embed: Training a Reproducible Long Context Text Embedder. TMLR

  37. [37]

    Arnold Overwijk, Chenyan Xiong, Xiao Liu, Cameron VandenBerg, and Jamie Callan. 2022. Clueweb22: 10 billion web documents with visual and semantic information. arXiv:2211.15848

    work page arXiv 2022

  38. [38]

    Yingqi Qu, Yuchen Ding, Jing Liu, Kai Liu, Ruiyang Ren, Wayne Xin Zhao, Daxi- ang Dong, Hua Wu, and Haifeng Wang. 2021. RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering. In Proc. of NAACL-HLT. 5835–5847

    2021

  39. [39]

    Corbin Rosset, Ho-Lam Chung, Guanghui Qin, Ethan Chau, Zhuo Feng, Ahmed Awadallah, Jennifer Neville, and Nikhil Rao. 2025. Researchy Questions: A Dataset of Multi-Perspective, Decompositional Questions for Deep Research. InProc. of SIGIR. 3712–3722

    2025

  40. [40]

    Chris Samarinas, Alexander Krubner, Alireza Salemi, Youngwoo Kim, and Hamed Zamani. 2025. Beyond Factual Accuracy: Evaluating Coverage of Diverse Factual Information in Long-form Text Generation. InFindings of ACL. 13468–13482

    2025

  41. [41]

    David P Sander and Laura Dietz. 2021. EXAM: How to evaluate retrieve-and- generate systems for users who do not (yet) know what they want.DESIRES, 136–146

    2021

  42. [42]

    Ivan Stelmakh, Yi Luan, Bhuwan Dhingra, and Ming-Wei Chang. 2022. ASQA: Factoid questions meet long-form answers. InProc. of EMNLP. 8273–8288

    2022

  43. [43]

    Haochen Tan, Zhijiang Guo, Zhan Shi, Lu Xu, Zhili Liu, Yunlong Feng, Xiaoguang Li, Yasheng Wang, Lifeng Shang, Qun Liu, and Linqi Song. 2024. ProxyQA: An alternative framework for evaluating long-form text generation with large language models. InProc. of ACL. 6806–6827

    2024

  44. [44]

    Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models. InProc. of NeurIPS

    2021

  45. [45]

    Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal

  46. [46]

    Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge- Intensive Multi-Step Questions. InProc. of ACL. 10014–10037

  47. [47]

    Ellen M Voorhees. 2003. Evaluating answers to definition questions. InProc. of NAACL-HLT

    2003

  48. [48]

    Zhichao Wang, Bin Bi, Yanqi Luo, Sitaram Asur, and Claire Na Cheng. 2025. Diversity enhances an LLM’s performance in RAG and long-context task. arXiv:2502.09017

    work page arXiv 2025

  49. [49]

    Benjamin Warner, Antoine Chaffin, Benjamin Clavié, Orion Weller, Oskar Hall- ström, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, Griffin Thomas Adams, Jeremy Howard, and Iacopo Poli. 2025. Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference. In...

    2025

  50. [50]

    Martin Wechsler and Peter Schäuble. 2000. The probability ranking principle revisited.Inf. Retr. Boston.3, 3 (2000), 217–227

    2000

  51. [51]

    Bennett, Junaid Ahmed, and Arnold Overwijk

    Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. 2021. Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval. InProc. of ICLR

    2021

  52. [52]

    Xiao Yang, Kai Sun, Hao Xin, Yushi Sun, Nikita Bhalla, Xiangsen Chen, Sajal Choudhary, Rongze Daniel Gui, Ziran Will Jiang, Ziyu Jiang, Lingkun Kong, Brian Moran, Jiaqi Wang, Yifan Ethan Xu, An Yan, Chenyu Yang, Eting Yuan, Hanwen Zha, Nan Tang, Lei Chen, Nicolas Scheffer, Yue Liu, Nirav Shah, Rakesh Wanga, Anuj Kumar, Wen-tau Yih, and Xin Luna Dong. 2024...

    2024

  53. [53]

    Wen-Tau Yih, Kristina Toutanova, John C Platt, and Christopher Meek. 2011. Learning Discriminative Projections for Text Similarity Measures. InProc. of CoNLL. 247–256

    2011

  54. [54]

    Puxuan Yu, Razieh Rahimi, Zhiqi Huang, and James Allan. 2023. Search Result Diversification Using Query Aspects as Bottlenecks. InProc. of CIKM. 3040–3051

    2023

  55. [55]

    Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou

  56. [56]

    Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models

    Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models. arXiv:2506.05176

    work page internal anchor Pith review Pith/arXiv arXiv

  57. [57]

    Yunfei Zhong, Jun Yang, Yixing Fan, Jiafeng Guo, Lixin Su, Maarten de Rijke, Ruqing Zhang, Dawei Yin, and Xueqi Cheng. 2025. Reasoning-enhanced query understanding through Decomposition and Interpretation. arXiv:2509.06544

    work page arXiv 2025