Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Fact-Preserved Personalized News Headline Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Personalized news headlines can stay faithful to the facts, not just to the reader's tastes, if user interests are tied to the candidate article and trained with contrastive examples.

desk verdict Useful architecture, shaky evaluation: the factual-consistency claim is built on an unvalidated FactCC loop and the personalization gains over BART are too small to call a balance. read the letter →

arxiv 2501.11828 v1 pith:QIUVTHWM submitted 2025-01-21 cs.CL cs.AI

Zhao Yang , Junhong Lian , Xiang Ao This is my paper · ORCID
classification cs.CLcs.AI
keywords personalizednewsheadlinegenerationfactualconsistencyuserembeddingcontrastivelearningencoder-decoderPENSbenchmarkabstractivesummarizationattention
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

The paper argues that personalized news headline generation does not have to sacrifice factual consistency for user appeal. It proposes FPG, which uses the similarity between a candidate news article and a user's clicked history to weight which facts matter, producing a fact-aware user embedding that guides decoding. A contrastive training stage then pulls generated headlines toward factually consistent examples and away from deliberately distorted ones. On the PENS benchmark, FPG reports the highest coverage, factuality, and personalization scores among the compared systems, suggesting the tension assumed by earlier work is not forced.

What carries the argument

The central object is the fact-aware global user embedding: a weighted sum of historical clicked-headline representations, where the weights come from a history-cross attention sub-layer that measures similarity between the candidate news body and each clicked headline. This embedding replaces the standard beginning-of-sentence token in the decoder, so it steers every decoding step toward user-relevant facts without injecting arbitrary user preferences. The second piece of machinery is a fact-enhanced contrastive training loss that uses FactCC-ranked positive examples and rule-based negative examples to teach the model to prefer factually consistent continuations.

What would settle it

A human-annotation study on a sample of generated headlines could settle the claim: take roughly 200 headlines from FPG and BART on the PENS test set, have three annotators mark factual errors, and compare. If FPG does not beat BART on human-judged faithfulness, the FactCC improvement does not reflect genuine factual consistency.

Watch

Extended reading notes

Core claim

The paper claims that blindly injecting a global user interest embedding into a headline decoder causes factual drift, especially when none of the user's historical clicks relate to the candidate news. FPG replaces that with a fact-aware global user embedding computed as a similarity-weighted sum of clicked-news representations, and it adds a contrastive learning phase that trains the generator toward factually consistent headlines and away from rule-based factual corruptions. In the reported experiments, FPG-GRU attains a FactCC score of 89.26, ROUGE-1 of 27.33, and personalization score of 2.88, compared with BART's 86.67, 26.27, and 2.72, which the paper reads as evidence that personalization and factual consistency can be pursued together.

Load-bearing premise

The paper assumes that the automatic factuality score it uses, FactCC, genuinely measures whether a headline states the article's facts rather than merely resembling the source text; this same score picks the positive training examples and is the reported consistency metric, so if it can be fooled by word overlap, the consistency gains may be an artifact of training to the metric.

Editorial extensions

If this is right

  • Factual consistency can be injected at the decoder's input side and through training objectives, rather than only through post-hoc filtering or reranking.
  • A user-conditioned headline generator can preserve source facts while still producing distinct headlines for different readers, because the user signal is aligned to the candidate article's content.
  • The contrastive training recipe could generalize to other user-conditioned text generation tasks where global user vectors are known to contaminate factual content.
  • The similarity-weighted attention gives the model a natural way to down-weight irrelevant historical clicks, reducing the risk of injecting misinformation from unrelated interests.

Reading between the lines

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

  • If FactCC can be gamed by surface overlap, part of the reported consistency gain may be metric optimization rather than genuine faithfulness; a human-annotation or question-answering factuality probe would separate the two.
  • A clean ablation would replace the learned similarity weights with uniform or random weights while keeping the contrastive stage; if FactCC stays high, the contrastive loss may be carrying most of the consistency improvement.
  • The same fact-aware user embedding idea could transfer to personalized summarization of financial or medical content, where a false entity association is costlier than in entertainment news.
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 / 5 minor

Summary. The paper proposes FPG (Fact-Preserved Personalized News Headline Generation), an encoder-decoder framework for personalized news headline generation. FPG augments a BART-initialized Transformer with a GRU-based history encoder, a personalized news encoder that uses history-cross attention between the candidate news body and the user's clicked headlines, and a user-guided decoder that replaces the [BOS] token with a fact-aware global user embedding. A contrastive fact-enhanced training stage is added to improve factual consistency. Experiments on the PENS benchmark report that FPG-GRU outperforms personalized and non-personalized baselines on ROUGE-1/2/L, FactCC, and the paper's personalization metric PC(avg)/PC(max), with a case study illustrating the desired behavior.

Significance. The task is timely and practically important: personalized headline generation must balance user appeal with faithfulness to the source article. The proposed architecture is clean and the use of a public benchmark (PENS) is a strength. If the factual-consistency gains were independently verified, the paper would make a useful contribution. However, the central claim currently rests on FactCC, a weakly supervised model-based scorer whose validity for this task is not established, and the personalization gains over the non-personalized BART baseline are numerically small. The paper also reports no significance tests despite claiming 'significantly outperforming.' These issues limit the strength of the empirical conclusions but are addressable with additional evaluation.

major comments (4)
  1. [Section V-C and Table II] The factual-consistency evaluation relies entirely on FactCC, a weakly supervised model-based scorer, with no validation of its reliability for generated news headlines. The paper provides no correlation of FactCC with human judgments, no human annotation of outputs, and no independent entailment- or QA-based metric. Since the central claim is improved factual consistency, an independent evaluation is required; otherwise the reported FactCC advantage of FPG-GRU (89.26 vs. 86.67 for BART) cannot be interpreted as greater faithfulness.
  2. [Section IV-D and Section V-C] The contrastive training set D* is constructed by selecting 'prominently ranked headline samples with high factual accuracy scores' without specifying the scoring method, while the evaluation uses FactCC, a model-based consistency scorer. If the same scorer family is used both to select positive examples and to score generated outputs, the reported FactCC gain partly reflects training to the metric. Please specify the exact scoring procedure used to build D*, and either use gold/original headlines as positives or verify the positive and negative examples with human judgments.
  3. [Table II and Section VI-A] The text claims that FPG-GRU 'significantly outperforms' baselines, but only single-run results are reported, with no error bars, no multiple seeds, and no significance tests. This is particularly important for the small gaps in ROUGE-1 (27.33 vs. 26.27 for BART) and FactCC (89.26 vs. 86.67). Please report variance across at least three runs and provide paired significance tests, or soften the significance claim accordingly.
  4. [Section V-C, Eq. (9), and Table II] The personalization metric PC(avg)/PC(max) measures cosine similarity between the generated headline and the user's historically clicked headlines. This metric can reward overlap with generic or repeated vocabulary and does not directly measure user-specific preference. More importantly, FPG-GRU's PC(avg)/PC(max) values (2.88/17.27) are close to the non-personalized BART baseline (2.72/17.13) and much lower than the personalized PENS-NAML baseline (3.93/22.73). The claim of striking a 'balance' needs additional supporting evidence, such as human evaluation of personalization or agreement with annotator-written personalized headlines, since on the reported metric FPG appears closer to BART than to personalized baselines.
minor comments (5)
  1. [Algorithm 1] The line 'Froze ξ to train the history encoder' is ambiguous because the subsequent update step minimizes Eq. (7) over θ; please state explicitly which parameter groups are updated in each of the four stages.
  2. [Section IV-D] The rule-based methods used to create negative examples are not enumerated; please specify them (e.g., entity substitution, numeric changes, negation) and state how many negatives are paired with each positive.
  3. [Section V-D] The text says 'The epoch number for the pre-trained phrase is 5'; this should read 'pre-trained phase,' and the list of learning rates should be explicitly mapped to the four training stages.
  4. [Section V-A and Table I] The training sets D10 and D3 appear in Table I but are not described in the text; please clarify how the limit l=10 and l=3 are used and why D5 is preferred for the main experiments.
  5. [Algorithm 1 and Section IV-D] The notation D* is used in Algorithm 1 before it is defined in Section IV-D; consider reordering or adding a forward reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No demonstrated circularity; the FactCC-based training/eval overlap is an unverified risk, not a shown equivalence.

full rationale

The paper's derivation chain is self-contained and does not reduce to its inputs by construction. The personalized news encoder (Eq. 3–4) and user-guided decoder (Eq. 5–6) are defined from the candidate news and the user's historical clicks, not from the evaluation metrics. The personalization metric PC (Eq. 9) is an external cosine-similarity measure, and the model is trained via MLE (Eq. 7) and contrastive learning (Eq. 8), not by directly optimizing PC, so the personalization result has independent content: non-personalized BART scores lower on PC despite also conditioning on the source text. The only plausible circularity is the factual-consistency path: Section IV-D constructs the contrastive set D* using 'high factual accuracy scores' while Section V-C evaluates with FactCC. However, the paper never states that the D* selector is FactCC; it says only 'We selected the prominently ranked headline samples with high factual accuracy scores compared to the news articles as positive instances.' Without an exhibited identity between the training selector and the evaluation metric, this is a validation and metric-choice risk, not a demonstrated circular step. The self-citation to PENS [11], whose authors include the corresponding author, is ordinary benchmark usage: PENS is an external dataset with manually annotated headlines, and the paper's model is not defined in terms of PENS's metrics. No uniqueness theorem or ansatz is imported from the authors' prior work. The central limitation, that FactCC is weakly supervised and may be gameable, belongs to correctness risk rather than circularity.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The model is built from standard neural components and PENS data. The main hidden cost is the FactCC-based data selection, which is both a free threshold choice and an axiom about metric validity. No new physical or conceptual entities are introduced; the 'fact-aware global user embedding' is a weighted sum of existing click-headline representations.

free parameters (8)
  • Network parameters of history encoder, personalized news encoder, and user-guided decoder = Learned on PENS via Eqs. (7) and (8)
    All projection matrices (H_Q, H_K, H_V, V_a, b_a, W_v, b_v), GRU weights, and BART-base parameters are fine-tuned on PENS distant supervision; standard for neural models, listed for completeness.
  • Encoder and decoder block count N = 6
    Architecture depth chosen without ablation in the paper.
  • Embedding dimension d_e = 768
    Set to BART-base dimension, no independent justification.
  • Number of attention heads = 12
    Default BART-base configuration.
  • Learning rates for four training stages = 3e-5, 1e-4, 3e-5, 1e-7
    Hand-selected schedule; no sensitivity analysis.
  • Beam size at decoding = 3
    Decoding hyperparameter, no ablation.
  • Max users per news in distant supervision (D5) = 5
    Data filter to avoid broad-appeal news; D10 and D3 are defined but not compared in the main results table.
  • Epoch counts per stage = 5, 5, 9, 1
    Training schedule; no early-stopping criterion or sensitivity analysis is reported.
assumptions (6)
  • standard math Transformer and GRU computations as given in [13, 38] are adopted without modification.
    Backbone architecture is assumed valid.
  • domain assumption PENS historical clicked headlines are a valid proxy for user reading interests.
    The personalization signal is entirely derived from click sequences; if clicks are noisy or mixed, the learned user embedding is not fact-aware.
  • domain assumption The original headline of a news item is an acceptable distant-supervision label for personalized headline generation.
    Section V-A states original headlines are used as imperfect labels; this turns a recommendation task into a generation task.
  • domain assumption FactCC is a reliable oracle for factual consistency, both for selecting contrastive positives and for evaluation.
    Section IV-D selects D* positives by high factual accuracy scores; Section V-C uses FactCC to measure consistency. This assumption carries the circularity burden.
  • domain assumption Cosine similarity between generated headlines and historical clicked titles measures personalization.
    Eq. (9) defines personalization as similarity to the user's click sequence; this rewards style overlap and can be inflated by repetition.
  • ad hoc to paper News clicked by a maximum of five users provide better supervision than popular news.
    Section V-A limits each news to at most 5 users without showing that D5 is better than D10 or D3 on the final metrics; this is a modeling choice specific to this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fact-Preserved Personalized News Headline Generation." pith.science (2026). https://pith.science/paper/QIUVTHWM

@misc{pith2026250111828,
  author       = {Pith},
  title        = {Pith review of: Fact-Preserved Personalized News Headline Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QIUVTHWM}},
  note         = {Machine review of arXiv:2501.11828}
}
read the original abstract

Personalized news headline generation, aiming at generating user-specific headlines based on readers' preferences, burgeons a recent flourishing research direction. Existing studies generally inject a user interest embedding into an encoderdecoder headline generator to make the output personalized, while the factual consistency of headlines is inadequate to be verified. In this paper, we propose a framework Fact-Preserved Personalized News Headline Generation (short for FPG), to prompt a tradeoff between personalization and consistency. In FPG, the similarity between the candidate news to be exposed and the historical clicked news is used to give different levels of attention to key facts in the candidate news, and the similarity scores help to learn a fact-aware global user embedding. Besides, an additional training procedure based on contrastive learning is devised to further enhance the factual consistency of generated headlines. Extensive experiments conducted on a real-world benchmark PENS validate the superiority of FPG, especially on the tradeoff between personalization and factual consistency.

Figures

Figures reproduced from arXiv: 2501.11828 by the authors.

Figure 1
Figure 1. An example illustrating the personalization in news headlines. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The framework of FPG. It has N layers of transformer blocks in both news encoder and decoder. (a) is history encoder, (b) is personalized news encoder, and (c) is user-guided decoder. historical clicks, i.e., Cu, and the news body of v, i.e., Xv, where y u j ∈ V for all 1 ≤ j ≤ T. IV. METHODOLOGY This section details our proposed FPG model, which is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 45 canonical work pages

  1. [1]

    Hedge trimmer: A parse-and-trim approach to headline generation,

    B. Dorr, D. Zajic, and R. Schwartz, “Hedge trimmer: A parse-and-trim approach to headline generation,” in Proceedings of the HLT-NAACL 03 Text Summarization Workshop, 2003

    work page 2003

  2. [2]

    HEADY: News headline abstraction through event pattern clustering,

    E. Alfonseca, D. Pighin, and G. Garrido, “HEADY: News headline abstraction through event pattern clustering,” in Proceedings of ACL , 2013

    work page 2013

  3. [3]

    Generating news headlines with recurrent neural net- works,

    K. Lopyrev, “Generating news headlines with recurrent neural net- works,” arXiv preprint arXiv:1512.01712 , 2015

    arXiv 2015

  4. [4]

    Neural headline generation on Abstract Meaning Representation,

    S. Takase, J. Suzuki, N. Okazaki, T. Hirao, and M. Nagata, “Neural headline generation on Abstract Meaning Representation,” in Proceed- ings of EMNLP , 2016

    work page 2016

  5. [5]

    From neural sentence summarization to headline generation: A coarse-to-fine approach,

    J. Tan, X. Wan, and J. Xiao, “From neural sentence summarization to headline generation: A coarse-to-fine approach,” in Proceedings of IJCAI, 2017

    work page 2017

  6. [6]

    Reading like HER: Human reading inspired extractive summarization,

    L. Luo, X. Ao, Y . Song, F. Pan, M. Yang, and Q. He, “Reading like HER: Human reading inspired extractive summarization,” in Proceedings of EMNLP, 2019

    work page 2019

  7. [7]

    Self-attentive model for headline generation,

    D. Gavrilov, P. Kalaidin, and V . Malykh, “Self-attentive model for headline generation,” in Proceedings of ECIR , 2019

    work page 2019

  8. [8]

    Generating representative headlines for news stories,

    X. Gu, Y . Mao, J. Han, J. Liu, Y . Wu, C. Yu, D. Finnie, H. Yu, J. Zhai, and N. Zukoski, “Generating representative headlines for news stories,” in Proceedings of WWW , 2020

    work page 2020

Show all 46 references
  1. [9]

    Pegasus: Pre-training with extracted gap-sentences for abstractive summarization,

    J. Zhang, Y . Zhao, M. Saleh, and P. J. Liu, “Pegasus: Pre-training with extracted gap-sentences for abstractive summarization,” in Proceedings of ICML, 2020

    2020

  2. [10]

    Few-shot text generation with natural language instructions,

    T. Schick and H. Sch ¨utze, “Few-shot text generation with natural language instructions,” in Proceedings of EMNLP , 2021

    2021

  3. [11]

    PENS: A dataset and generic framework for personalized news headline generation,

    X. Ao, X. Wang, L. Luo, Y . Qiao, Q. He, and X. Xie, “PENS: A dataset and generic framework for personalized news headline generation,” in Proceedings of ACL , 2021

    2021

  4. [12]

    When framing matters: How partisan and journalistic frames affect individual opinions and party identification,

    M. W. Wagner and M. Gruszczynski, “When framing matters: How partisan and journalistic frames affect individual opinions and party identification,” Journalism & Communication Monographs , 2016

    2016

  5. [13]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of NIPS, 2017

    2017

  6. [14]

    Neural news recommendation with attentive multi-view learning,

    C. Wu, F. Wu, M. An, J. Huang, Y . Huang, and X. Xie, “Neural news recommendation with attentive multi-view learning,” in Proceedings of IJCAI, 2019

    2019

  7. [15]

    Neural news recom- mendation with multi-head self-attention,

    C. Wu, F. Wu, S. Ge, T. Qi, Y . Huang, and X. Xie, “Neural news recom- mendation with multi-head self-attention,” in Proceedings of EMNLP , 2019

    2019

  8. [16]

    Npa: Neural news recommendation with personalized attention,

    C. Wu, F. Wu, M. An, J. Huang, Y . Huang, and X. Xie, “Npa: Neural news recommendation with personalized attention,” in Proceedings of KDD, 2019

    2019

  9. [17]

    Neural news recommendation with long- and short-term user representations,

    M. An, F. Wu, C. Wu, K. Zhang, Z. Liu, and X. Xie, “Neural news recommendation with long- and short-term user representations,” in Proceedings of ACL , 2019

    2019

  10. [18]

    Supervised contrastive learn- ing,

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learn- ing,” in Proceedings of NeurIPS , 2020

    2020

  11. [19]

    Supervised contrastive learning for pre-trained language model fine-tuning,

    B. Gunel, J. Du, A. Conneau, and V . Stoyanov, “Supervised contrastive learning for pre-trained language model fine-tuning,” in Proceedings of ICLR, 2021

    2021

  12. [20]

    Improving factual consistency of abstractive summarization via question answering,

    F. Nan, C. Nogueira dos Santos, H. Zhu, P. Ng, K. McKeown, R. Nallap- ati, D. Zhang, Z. Wang, A. O. Arnold, and B. Xiang, “Improving factual consistency of abstractive summarization via question answering,” in Proceedings of ACL, 2021

    2021

  13. [21]

    Event-driven headline gener- ation,

    R. Sun, Y . Zhang, M. Zhang, and D. Ji, “Event-driven headline gener- ation,” in Proceedings of ACL , 2015

    2015

  14. [22]

    Get to the point: Summarization with pointer-generator networks,

    A. See, P. J. Liu, and C. D. Manning, “Get to the point: Summarization with pointer-generator networks,” in Proceedings of ACL , 2017

    2017

  15. [23]

    BERT: Pre- training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- training of deep bidirectional transformers for language understanding,” in Proceedings of NAACL, 2019

    2019

  16. [24]

    Exploring the limits of transfer learning with a unified text-to-text transformer,

    C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y . Zhou, W. Li, and P. J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer,” Journal of Machine Learning Research, 2020

    2020

  17. [25]

    BART: Denoising sequence-to- sequence pre-training for natural language generation, translation, and comprehension,

    M. Lewis, Y . Liu, N. Goyal, M. Ghazvininejad, A. Mohamed, O. Levy, V . Stoyanov, and L. Zettlemoyer, “BART: Denoising sequence-to- sequence pre-training for natural language generation, translation, and comprehension,” in Proceedings of ACL , 2020

    2020

  18. [26]

    BRIO: Bringing order to abstractive summarization,

    Y . Liu, P. Liu, D. Radev, and G. Neubig, “BRIO: Bringing order to abstractive summarization,” in Proceedings of ACL, 2022

    2022

  19. [27]

    News headline generation based on improved decoder from transformer,

    Z. Li, J. Wu, J. Miao, and X. Yu, “News headline generation based on improved decoder from transformer,” Scientific Reports, 2022

    2022

  20. [28]

    Style transfer from non- parallel text by cross-alignment,

    T. Shen, T. Lei, R. Barzilay, and T. Jaakkola, “Style transfer from non- parallel text by cross-alignment,” in Proceedings of NIPS , 2017

    2017

  21. [29]

    Style transfer in text: Exploration and evaluation,

    Z. Fu, X. Tan, N. Peng, D. Zhao, and R. Yan, “Style transfer in text: Exploration and evaluation,” in Proceedings of AAAI , 2018

    2018

  22. [30]

    Style transfer through back-translation,

    S. Prabhumoye, Y . Tsvetkov, R. Salakhutdinov, and A. W. Black, “Style transfer through back-translation,” in Proceedings of ACL , 2018

    2018

  23. [31]

    Deep headline generation for clickbait detection,

    K. Shu, S. Wang, T. Le, D. Lee, and H. Liu, “Deep headline generation for clickbait detection,” in Proceedings of ICDM , 2018

    2018

  24. [32]

    Question headline generation for news articles,

    R. Zhang, J. Guo, Y . Fan, Y . Lan, J. Xu, H. Cao, and X. Cheng, “Question headline generation for news articles,” in Proceedings of CIKM, 2018

    2018

  25. [33]

    Clickbait? sensational head- line generation with auto-tuned reinforcement learning,

    P. Xu, C.-S. Wu, A. Madotto, and P. Fung, “Clickbait? sensational head- line generation with auto-tuned reinforcement learning,” in Proceedings of EMNLP, 2019

    2019

  26. [34]

    Contrastive learning enhanced author-style headline generation,

    H. Liu, W. Guo, Y . Chen, and X. Li, “Contrastive learning enhanced author-style headline generation,” in Proceedings of EMNLP , 2022

    2022

  27. [35]

    Transformer reasoning network for personalized review summarization,

    H. Xu, H. Liu, P. Jiao, and W. Wang, “Transformer reasoning network for personalized review summarization,” in Proceedings of SIGIR, 2021

    2021

  28. [36]

    Rein- forcing pretrained models for generating attractive text advertisements,

    X. Wang, X. Gu, J. Cao, Z. Zhao, Y . Yan, B. Middha, and X. Xie, “Rein- forcing pretrained models for generating attractive text advertisements,” in Proceedings of KDD , 2021

    2021

  29. [37]

    Personalized headline generation with enhanced user interest perception,

    K. Zhang, G. Lu, G. Zhang, Z. Lei, and L. Wu, “Personalized headline generation with enhanced user interest perception,” in Proceedings of ICANN, 2022

    2022

  30. [38]

    Learning phrase representations using RNN encoder–decoder for statistical machine translation,

    K. Cho, B. van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio, “Learning phrase representations using RNN encoder–decoder for statistical machine translation,” in Proceed- ings of EMNLP , 2014

    2014

  31. [39]

    Improving truthfulness of headline generation,

    K. Matsumaru, S. Takase, and N. Okazaki, “Improving truthfulness of headline generation,” in Proceedings of ACL, 2020

    2020

  32. [40]

    Rdf-to-text generation with graph- augmented structural neural encoders,

    H. Gao, L. Wu, P. Hu, and F. Xu, “Rdf-to-text generation with graph- augmented structural neural encoders,” in Proceedings of IJCAI , 2020

    2020

  33. [41]

    CLIFF: Contrastive learning for improving faithfulness and factuality in abstractive summarization,

    S. Cao and L. Wang, “CLIFF: Contrastive learning for improving faithfulness and factuality in abstractive summarization,” in Proceedings of EMNLP, 2021

    2021

  34. [42]

    Searching for effective neural extractive summarization: What works and what’s next,

    M. Zhong, P. Liu, D. Wang, X. Qiu, and X. Huang, “Searching for effective neural extractive summarization: What works and what’s next,” in Proceedings of ACL , 2019

    2019

  35. [43]

    Embedding-based news recommendation for millions of users,

    S. Okura, Y . Tagami, S. Ono, and A. Tajima, “Embedding-based news recommendation for millions of users,” in Proceedings of KDD , 2017

    2017

  36. [44]

    ROUGE: A package for automatic evaluation of summaries,

    C.-Y . Lin, “ROUGE: A package for automatic evaluation of summaries,” in Text Summarization Branches Out , 2004

    2004

  37. [45]

    Evaluating the factual consistency of abstractive text summarization,

    W. Kryscinski, B. McCann, C. Xiong, and R. Socher, “Evaluating the factual consistency of abstractive text summarization,” in Proceedings of EMNLP, 2020

    2020

  38. [46]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proceedings of ICLR , 2019

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.