Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Beyond Quality: Unlocking Diversity in Ad Headline Generation with Large Language Models

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims DIVER, a two-stage framework with synthetic data and multi-objective reinforcement learning, generates diverse ad headlines in one pass and, deployed at scale, raises advertiser value by 4.0% and click-through rate by 1.4%

desk verdict A solid industrial pipeline for generating diverse ad headlines, but the online A/B claim is under-specified: the serving-time selector may explain the lift, so take the +4.0% ADVV with caution. read the letter →

arxiv 2508.18739 v1 pith:3IL7K6HD submitted 2025-08-26 cs.CL cs.LG

Chang Wang , Siyu Yan , Depeng Yuan , Yuqi Chen , Yanhua Huang , Yuanhang Zheng , Shuhao Li , Yinqi Zhang
show 3 more authors
Kedi Chen Mingrui Zhu Ruiwen Xu
This is my paper
classification cs.CLcs.LG
keywords adheadlinegenerationdiversitylargelanguagemodelsreinforcementlearningmulti-objectiveoptimizationGRPOadvertisingpersonalization
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 show that ad headline diversity can be treated as an explicit training objective rather than a by-product of sampling. It argues that two interventions—a synthetic data pipeline that annotates headlines with semantic keywords and stylistic types, and a reinforcement-learning stage that rewards low pairwise BLEU, style coverage, faithfulness, predicted click-through rate, correct count, and parseable format—let one LLM emit many diverse, high-quality headlines in one pass. Offline, the method beats closed- and open-weight baselines on diversity while staying near the best on quality. Online, on a large content-sharing platform, the deployed system improves advertiser value by 4.0% and click-through rate by 1.4% against a sampling-plus-supervised-fine-tuning baseline.

What carries the argument

The mechanism is the set-level composite reward plus the two-stage training that makes it usable. Pairwise BLEU and style coverage are combined into a diversity score; a binary faithfulness classifier and a click-through prediction model supply quality and user-preference scores; quantity and JSON-format rules keep the output usable. Averaging these five rewards and optimizing with GRPO, a policy-gradient reinforcement-learning algorithm, pushes the model to generate N headlines that are simultaneously different, on-topic, engaging, and parseable, in one forward pass.

What would settle it

An A/B test that keeps the serving-time user-profile selection step but replaces the reinforcement-learning-trained generator with the supervised-fine-tuning-only checkpoint from the paper's ablations: if the 4.0% advertiser-value and 1.4% click-through-rate gains disappear, the reinforcement-learning stage is the cause; if they persist, the gains come from the synthetic data or the selector.

Watch

Extended reading notes

Core claim

The paper's central claim is that diversity in one-to-many generation is not a decoding-time accident; it can be optimized at the set level. DIVER first builds synthetic multi-headline training data by extracting a semantic keyword and a style (directness, emoji usage, rhetorical device) from real ad headlines, then fine-tunes a generator to follow those cues, then applies GRPO to maximize a composite reward over the whole generated set. The reward averages five signals—diversity (1 minus average pairwise BLEU combined with style coverage), faithfulness to the ad content, predicted click-through rate, output count against target, and JSON parseability. The paper reports that this produces th

Load-bearing premise

The paper's results rest on the assumption that the five automated reward scores—especially the click-through model trained on the best and worst supervised-fine-tuning outputs—are faithful proxies for what makes an ad headline valuable; if that model is biased, the measured quality and the online gains could come from the serving-time user-profile matching rather than from the generation training.

Editorial extensions

If this is right

  • A single forward pass can replace sampling-based diversity tricks: the model is trained to emit N diverse headlines at once, so serving needs no rejection sampling or prompt reshuffling.
  • Diversity and quality are not necessarily in tension: the composite reward finds a point where pairwise BLEU drops to 2.08 while NLI faithfulness stays near the best baselines.
  • Both stages carry weight: ablations show removing the synthetic data pipeline or the reinforcement-learning stage separately degrades diversity and faithfulness, and removing both is worst.
  • The gains materialize in business metrics: deployed with user-profile matching over 30 candidates, the system improves advertiser value by 4.0%, click-through rate by 1.4%, impressions by 2.4%, and cost per mile by 2.0%.

Reading between the lines

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

  • The same two-stage recipe is likely portable to other one-to-many creative generation tasks, such as product descriptions, social post variants, or email subject lines, because nothing in the reward design is ad-specific except the click-through model.
  • The pairwise-BLEU part of the diversity reward may under-reward legitimate reuse of key product terms; embedding-based set diversity could capture semantic spread without penalizing those words.
  • The serving-time user-profile matcher is exactly where diversity and personalization meet: more diverse candidate sets give the matcher more room, so a sweep over candidate count could isolate how much of the online gain comes from training quality versus selection quality.
  • The online numbers conflate the trained generator with the serving selector; applying the same user-profile matcher to a stronger supervised-fine-tuning-only generator would pinpoint the marginal value of the reinforcement-learning stage.
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 proposes DIVER, a multi-stage framework for ad headline generation that combines a semantic-/stylistic-aware synthetic data pipeline (for cold-start SFT) with multi-objective reinforcement learning (GRPO) using five reward components: diversity, quality, CTR, quantity, and format. The authors report offline results on a 3,000-instance industrial test set showing lower PairBLEU/SelfBLEU and higher StyleCov than strong LLM baselines while maintaining NLI/ROUGE quality, plus ablations indicating each component matters. They also report an online A/B deployment on Xiaohongshu's Explore Feed, claiming +4.0% ADVV and +1.4% CTR relative to a 'Sampling + SFT' baseline, with the serving-time step of selecting the generated headline most semantically similar to the user profile.

Significance. If the claims hold, DIVER would be a useful industrial contribution: it shows that a diversity-aware reward can be combined with quality and CTR objectives without sacrificing faithfulness, and it is deployed at scale. The paper gives a concrete, reproducible-looking recipe (data augmentation, SFT, GRPO with five rewards) and includes thorough ablations that demonstrate each reward and each pipeline stage has a measurable effect. However, the two load-bearing empirical claims—offline diversity/quality balance and online business lift—are weakened by (i) direct overlap between the diversity reward and the offline diversity metrics, and (ii) an under-specified online A/B contrast that does not isolate generation quality from the serving-time selection mechanism. The lack of variance reporting and the single-split test further limit confidence in the precision of the reported numbers.

major comments (4)
  1. [§4.6, Table 4] The online A/B test is under-specified. The text says DIVER generates 30 headlines and then selects 'the headline most semantically similar to the user profile.' For the 'Sampling + SFT' baseline, it is not stated whether the same 30-headline generation procedure, the same similarity metric, and the same user-profile representation are used. If the baseline serves without user-profile matching (or with a different selector), the reported +4.0% ADVV and +1.4% CTR may be driven by the selection step, not by the SFT+RL generation pipeline that is the paper's central contribution. Please specify the baseline serving protocol explicitly, and ideally include a third arm that uses the same selector on the baseline model's outputs.
  2. [§3.2.1 vs §4.1, Tables 1–2] There is a direct overlap between the diversity reward and the offline evaluation metrics. The reward is rdiversity = (1 − PairBLEU(Y) + Coverage(Y))/2, while Table 1's diversity columns include PairBLEU and StyleCov (and SelfBLEU/DisNGram/CosSim are correlated). Thus the offline diversity gains are partially by construction; the model is optimized against the same statistics that are later used to claim success. The paper should add an independent diversity evaluation (e.g., human judgments of distinctiveness, or a downstream task where diversity is not part of the training objective) or explicitly acknowledge that the offline diversity metrics are not independent of the reward.
  3. [§4.2–§4.3, Tables 1–2] No variance or significance testing is reported anywhere. All offline results are point estimates on a single 3,000-instance test split. For a paper making an industrial deployment claim, it is essential to report confidence intervals or at least significance tests (e.g., bootstrap or multiple evaluation seeds) for the main tables. This is not a perfunctory request: the differences in some quality columns (e.g., Rouge-1: 16.71 vs 17.93 for Qwen2.5-72B in Table 1) are small and may be within noise, which would affect the 'maintains high quality' claim.
  4. [Appendix A, 'Details about CTR Reward'] The CTR reward model is trained on headlines generated by high-temperature SFT, labeled by the top and bottom third of actual CTR. During RL, the model being optimized shifts the distribution of generated headlines; there is no evidence that the CTR predictor remains calibrated or discriminative on DIVER-generated headlines. Since the online CTR lift is a headline result, the paper should report some validation of the CTR reward model on held-out head-to-head data, or at least discuss the distribution shift. Without this, the mechanism by which GRPO improves real CTR is unclear.
minor comments (4)
  1. [§3.2.1, Eq. (1)] The normalization of PairBLEU and Coverage in rdiversity is ambiguous. Table 1 reports PairBLEU values like 2.08 and StyleCov as percentages (63.42%). If these are on different scales (0–100 vs 0–1), the arithmetic in Eq. (1) is undefined. Please state the assumed scales and, if necessary, normalize explicitly.
  2. [Ethical Considerations] Typo: 'All datasets used in this study used are properly licensed' should be 'All datasets used in this study are properly licensed.'
  3. [§4.1, Evaluation Metrics] The metric names 'Rouge-1', 'Rouge-2', 'Rouge-L' should be capitalized as 'ROUGE-1', 'ROUGE-2', 'ROUGE-L' to match standard usage (and the reference 'Chin-Yew, 2004' is also typically 'Lin, 2004').
  4. [Table 3] In the 'Home Improvement Fence Ad' column, 'User Type 1' is described as 'Male DIY & fitness lover'—'fitness lover' is not obviously relevant to a fence-installation ad; consider whether this is a typo (e.g., 'DIY & home-improvement lover').

Circularity Check

1 steps flagged · score 5.0 of 10

Diversity evaluation metrics are identical to the RL diversity reward; offline diversity gains are partly by construction, though quality and online metrics remain independent.

  1. self definitional [Section 3.2.1 (rdiversity) vs Section 4.1 / Table 1; Appendix A]
    "Sec. 3.2.1: 'rdiversity = 1 − Pair-BLEU(Y ) + Coverage(Y ) 2' ... Sec. 4.1: 'To assess diversity, we measure ... Pairwise BLEU ... We evaluate style diversity via Style Coverages.' App. A: 'Style diversity is measured by the proportion of distinct style categories presents in the generated headlines, where we prompt DeepSeek-V3 to classify the style of the headline.'"

    The diversity reward maximizes (1 - PairBLEU + Coverage)/2, and Table 1's diversity columns include PairBLEU and StyleCov computed with the same 16-style taxonomy and DeepSeek-V3 style classifier. GRPO directly optimizes these quantities, so the claimed PairBLEU/StyleCov improvements for DIVER are consequences of maximizing the reward, not independent verification of diversity. The remaining diversity metrics (SelfBLEU, DisNGram, CosSim) are not in the reward and provide partial independent evidence, but the headline claim that DIVER 'balances quality and diversity' leans heavily on the self-defined metrics.

full rationale

The paper has a genuine circular component: the diversity construct is defined by the same PairBLEU and style-coverage statistics that the RL reward optimizes, so the offline diversity result is partly a self-consistency check of the optimizer. This is not the whole paper—NLI/ROUGE quality metrics are not the same as the quality reward (which is a human-labeled classifier), and the online ADVV/CTR A/B results are at least in principle independent of the reward definition. I do not see load-bearing self-citation: the cited in-house works (Chen et al. 2025; Huang et al. 2025) are used for background, not to justify the core claim. The online A/B contrast is under-specified (the 'Sampling + SFT' baseline may not use the same 30-headline + profile-matching selector), but that is an experimental-control concern, not a circularity. The Limitations section also concedes that 'fixed reward metrics may overlook nuanced user preferences,' which reinforces the concern but does not remove the circularity. Overall, one central diversity-related step reduces by construction, so score 5.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central claim relies on a chain of learned proxies: style classifier (DeepSeek-V3) for coverage, quality classifier for faithfulness, CTR model for engagement, and cosine similarity for serving. The two reward models are fitted to data derived from the SFT model itself and from historical logs; their accuracy is not evaluated against the offline or online metrics separately.

free parameters (6)
  • Equal reward weights across five components = 1/5 each (implied by 'averaged across five components')
    Section 3.2.1: 'the overall reward averaged across five components'; no tuning analysis or sensitivity reported, yet the balance between diversity, quality, CTR, quantity, and format determines deployed behavior.
  • Faithfulness threshold for quality reward = unspecified
    Section 3.2.1: 'The reward reflects the proportion of headlines that meet or exceed a given faithfulness threshold.' The threshold is not given, so the quality reward is not fully specified.
  • CTR pairwise margin = 0.3
    Appendix A: L = max(0, 0.3 - s+ + s-); the margin is hand-set and part of the CTR reward model training.
  • CTR pair sampling split (top/bottom third) = top and bottom third of headlines by CTR
    Appendix A: 'user interaction data is used to label the top and bottom third of headlines by CTR as positive and negative samples'; this data selection defines the CTR reward model's training signal.
  • Number of generated headlines per content (N) = 30 in online serving; target T in training unspecified
    Section 4.6: 'we first generate 30 ad headlines with DIVER'; the quantity reward r_quantity = min(1, N/T) depends on the chosen target T, and the diversity/coverage metrics depend on N.
  • Style taxonomy discretization = 2 x 2 x 4 = 16 style types
    Section 3.1 footnote: directness (2), emoji (2), rhetorical type (4); the coverage reward and StyleCov evaluation score against this hand-chosen label space, which is not validated against user perception.
assumptions (5)
  • ad hoc to paper The 16-way style taxonomy (directness x emoji x rhetorical type) decomposes headline diversity in a way that matches user engagement value.
    Section 3.1 footnote 2; the style diversity reward and StyleCov evaluation both assume this discretization is meaningful; no human validation of the taxonomy is reported.
  • domain assumption The CTR reward model trained on high-temperature SFT-generated headline pairs is a valid proxy for real CTR, and optimizing against it transfers online.
    Appendix A 'Details about CTR Reward': pairs come from SFT generations over 10,000 notes; RL then maximizes the proxy. Online A/B suggests transfer, but no offline-to-online correlation is shown.
  • ad hoc to paper Semantic similarity between a generated headline and the user profile is a sufficient personalization signal.
    Section 4.6: 'we select the headline most semantically similar to the user profile'; no validation or ablation of this selection rule is provided.
  • domain assumption NLI-based evaluation (mDeBERTa-v3 XNLI) and ROUGE scores adequately measure headline 'quality' for advertising.
    Section 4.1 evaluation metrics; no correlation with human quality judgments of advertisements is reported.
  • standard math GRPO policy optimization and the KL penalty to pi_sft behave as in Shao et al. (2024) for this multi-reward setup.
    Section 3.2.2 applies the GRPO objective; assumes standard convergence and that the composite reward is a valid scalarization.
invented entities (1)
  • 16-way ad headline style taxonomy (directness, emoji, rhetorical device)
    purpose: Defines the style space for the style diversity reward, data annotation, and evaluation coverage metric.
    No external evidence (user studies, advertiser surveys) that these 16 styles are the right decomposition of diversity or that high style coverage increases advertiser value.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Quality: Unlocking Diversity in Ad Headline Generation with Large Language Models." pith.science (2026). https://pith.science/paper/3IL7K6HD

@misc{pith2026250818739,
  author       = {Pith},
  title        = {Pith review of: Beyond Quality: Unlocking Diversity in Ad Headline Generation with Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IL7K6HD}},
  note         = {Machine review of arXiv:2508.18739}
}
read the original abstract

The generation of ad headlines plays a vital role in modern advertising, where both quality and diversity are essential to engage a broad range of audience segments. Current approaches primarily optimize language models for headline quality or click-through rates (CTR), often overlooking the need for diversity and resulting in homogeneous outputs. To address this limitation, we propose DIVER, a novel framework based on large language models (LLMs) that are jointly optimized for both diversity and quality. We first design a semantic- and stylistic-aware data generation pipeline that automatically produces high-quality training pairs with ad content and multiple diverse headlines. To achieve the goal of generating high-quality and diversified ad headlines within a single forward pass, we propose a multi-stage multi-objective optimization framework with supervised fine-tuning (SFT) and reinforcement learning (RL). Experiments on real-world industrial datasets demonstrate that DIVER effectively balances quality and diversity. Deployed on a large-scale content-sharing platform serving hundreds of millions of users, our framework improves advertiser value (ADVV) and CTR by 4.0% and 1.4%.

Figures

Figures reproduced from arXiv: 2508.18739 by the authors.

Figure 1
Figure 1. An illustration of diversified ad headline gen [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the DIVER framework. Our approach first performs synthetic data-augmented SFT to enable [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Prompts for the data enrichment. 10 [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Prompts for the data construction. 11 [PITH_FULL_IMAGE:figures/full_fig_p011_4.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.

  1. Design Your Ad: Personalized Advertising Image and Text Generation with Unified Autoregressive Models

    cs.CV 2026-05 unverdicted novelty 7.0 of 10

    Uni-AdGen uses a unified autoregressive framework with foreground perception, instruction tuning, and coarse-to-fine preference modules to generate personalized image-text ads from noisy user behaviors, outperforming ...

Reference graph

Works this paper leans on

39 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    arXiv 2023

  4. [4]

    Anthropic. 2024. https://www-cdn.anthropic.com/fed9cc193a14b84131812372d8d5857f8f304c52/Model_Card_Claude_3.pdf Claude 3.5 sonnet model card addendum

    work page 2024

  5. [5]

    Xiang Ao, Ling Luo, Xiting Wang, Zhao Yang, Jiun-Hung Chen, Ying Qiao, Qing He, and Xing Xie. 2023. Put your voice on stage: Personalized headline generation for news articles. ACM Trans. Knowl. Discov. Data, 18(3):20

    work page 2023

  6. [6]

    Xiang Ao, Xiting Wang, Ling Luo, Ying Qiao, Qing He, and Xing Xie. 2021. PENS : A dataset and generic framework for personalized news headline generation. 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), pages 82--9...

    work page 2021

  7. [7]

    Kevin Bartz, Cory Barr, and Adil Aijaz. 2008. Natural language generation for sponsored-search advertisements. In Proceedings of the 9th ACM Conference on Electronic Commerce, page 1–9

    work page 2008

  8. [8]

    Zheng Chai, Qin Ren, Xijun Xiao, Huizhi Yang, Bo Han, Sijun Zhang, Di Chen, Hui Lu, Wenlin Zhao, Lele Yu, et al. 2025. Longer: Scaling up long sequence modeling in industrial recommenders. arXiv preprint arXiv:2505.04421

    arXiv 2025

Show all 39 references
  1. [9]

    Kedi Chen, Qin Chen, Jie Zhou, Xinqi Tao, Bowen Ding, Jingwen Xie, Mingchen Xie, Peilong Li, and Zheng Feng. 2025. Enhancing uncertainty modeling with semantic graph for hallucination detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages...

    2025

  2. [10]

    Lin Chin-Yew. 2004. Rouge: A package for automatic evaluation of summaries. In Proceedings of the Workshop on Text Summarization Branches Out, 2004

    2004

  3. [11]

    Josef Dai, Xuehai Pan, Ruiyang Sun, Jiaming Ji, Xinbo Xu, Mickel Liu, Yizhou Wang, and Yaodong Yang. 2024. Safe rlhf: Safe reinforcement learning from human feedback. In The Twelfth International Conference on Learning Representations

    2024

  4. [13]

    Angela Fan, Mike Lewis, and Yann Dauphin. 2018. Hierarchical neural story generation. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 889--898

    2018

  5. [14]

    Atsushi Fujita, Katsuhiro Ikushima, Satoshi Sato, Ryo Kamite, Ko Ishiyama, and Osamu Tamachi. 2010. Automatic generation of listing ads by reusing promotional texts. In Proceedings of the 12th International Conference on Electronic Commerce: Roadmap for the Future of Electroni...

    2010

  6. [15]

    Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2020. The curious case of neural text degeneration. arXiv preprint arXiv:1904.09751

    2020 arXiv

  7. [16]

    Yanhua Huang, Yuqi Chen, Xiong Cao, Rui Yang, Mingliang Qi, Yinghao Zhu, Qingchang Han, Yaowei Liu, Zhaoyu Liu, Xuefeng Yao, et al. 2025. Towards large-scale generative ranking. arXiv preprint arXiv:2505.04180

    2025 arXiv

  8. [17]

    Yashal Shakti Kanungo, Sumit Negi, and Aruna Rajan. 2021. Ad headline generation using self-critical masked language model. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Indust...

    2021

  9. [18]

    Robert Kirk, Ishita Mediratta, Christoforos Nalmpantis, Jelena Luketina, Eric Hambro, Edward Grefenstette, and Roberta Raileanu. 2024. Understanding the effects of rlhf on llm generalisation and diversity. arXiv preprint arXiv:2310.06452

    2024 arXiv

  10. [19]

    Gregory Kang Ruey Lau, Wenyang Hu, Diwen Liu, Jizhuo Chen, See-Kiong Ng, and Bryan Kian Hsiang Low. 2024. Dipper: Diversity in prompts for producing large language model ensembles in reasoning tasks. arXiv preprint arXiv:2412.15238

    2024

  11. [20]

    Jiwei Li, Michel Galley, Chris Brockett, Jianfeng Gao, and Bill Dolan. 2015. A diversity-promoting objective function for neural conversation models. arXiv preprint arXiv:1510.03055

    2015 arXiv

  12. [21]

    Junhong Lian, Xiang Ao, Xinyu Liu, Yang Liu, and Qing He. 2025. Panoramic interests: Stylistic-content aware personalized headline generation. In Companion Proceedings of the ACM on Web Conference 2025, page 1109–1112

    2025

  13. [22]

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437

    2024 arXiv

  14. [23]

    Long Mai and Julie Carson-Berndsen. 2024. Improving linguistic diversity of large language models with possibility exploration fine-tuning. arXiv preprint arXiv:2412.03343

    2024 arXiv

  15. [24]

    Humza Naveed, Asad Ullah Khan, Shi Qiu, Muhammad Saqib, Saeed Anwar, Muhammad Usman, Naveed Akhtar, Nick Barnes, and Ajmal Mian. 2023. A comprehensive overview of large language models. arXiv preprint arXiv:2307.06435

    2023 arXiv

  16. [25]

    OpenAI. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276

    2024 arXiv

  17. [26]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, a...

    2022

  18. [27]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311--318

    2002

  19. [28]

    Qwen. 2025. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115

    2025 arXiv

  20. [29]

    Gerard Salton and Michael J. McGill. 1986. Introduction to Modern Information Retrieval. McGraw-Hill, Inc., USA

    1986

  21. [30]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300

    2024 arXiv

  22. [31]

    Yun-Zhu Song, Yi-Syuan Chen, Lu Wang, and Hong-Han Shuai. 2023. General then personal: Decoupling and pre-training for personalized headline generation. Transactions of the Association for Computational Linguistics, 11:1588--1607

    2023

  23. [32]

    Xiaoyu Tan, Leijun Cheng, Xihe Qiu, Shaojie Shi, Yuan Cheng, Wei Chu, Yinghui Xu, and Yuan Qi. 2024. Enhancing personalized headline generation via offline goal-conditioned reinforcement learning with large language models. In Proceedings of the 30th ACM SIGKDD Conference on K...

    2024

  24. [33]

    Guy Tevet and Jonathan Berant. 2021. Evaluating the evaluation of diversity in natural language generation. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 326--346, Online. Association for Comp...

    2021

  25. [34]

    Stamatina Thomaidou, Ismini Lourentzou, Panagiotis Katsivelis-Perakis, and Michalis Vazirgiannis. 2013. Automated snippet generation for online advertising. In Proceedings of the 22nd ACM International Conference on Information & Knowledge Management, page 1841–1844

    2013

  26. [35]

    Aditya Srinivas Timmaraju, Mehdi Mashayekhi, Mingliang Chen, Qi Zeng, Quintin Fettes, Wesley Cheung, Yihan Xiao, Manojkumar Rangasamy Kannadasan, Pushkar Tripathi, Sean Gahagan, et al. 2023. Towards fairness in personalized ads using impression variance aware reinforcement lea...

    2023

  27. [36]

    Zeqiu Wu, Yushi Hu, Weijia Shi, Nouha Dziri, Alane Suhr, Prithviraj Ammanabrolu, Noah A Smith, Mari Ostendorf, and Hannaneh Hajishirzi. 2023. Fine-grained human feedback gives better rewards for language model training. arXiv preprint arXiv:2306.01693

    2023 arXiv

  28. [37]

    Peng Xu, Chien-Sheng Wu, Andrea Madotto, and Pascale Fung. 2019. Clickbait? sensational headline generation with auto-tuned reinforcement learning. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confere...

    2019

  29. [38]

    Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan Berant. 2023. Making retrieval-augmented language models robust to irrelevant context

    2023

  30. [39]

    Kui Zhang, Guangquan Lu, Guixian Zhang, Zhi Lei, and Lijuan Wu. 2022. Personalized headline generation with enhanced user interest perception. In Artificial Neural Networks and Machine Learning – ICANN 2022: 31st International Conference on Artificial Neural Networks, Bristol,...

    2022

  31. [40]

    Yaoming Zhu, Sidi Lu, Lei Zheng, Jiaxian Guo, Weinan Zhang, Jun Wang, and Yong Yu. 2018. Texygen: A benchmarking platform for text generation models. In The 41st international ACM SIGIR conference on research & development in information retrieval, pages 1097--1100

    2018

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.