Pith. sign in

REVIEW 3 major objections 3 minor 46 references

IP2: Entity-Guided Interest Probing for Personalized News Recommendation

T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read IP2 claims that modeling entity-guided interest at two levels—which entity dominates headline scanning and how entity preference across reading history shapes title reading—makes a news recommender outperform all compared state-of-the-art…

desk verdict Solid news recommendation paper with a genuinely useful KG-free entity pretraining trick, but the behavioral 'probing' interpretation is not actually enforced by the loss. read the letter →

arxiv 2507.13622 v1 pith:VMVILIC7 submitted 2025-07-18 cs.IR

Youlin Wu , Yuanyuan Sun , Xiaokun Zhang , Haoxi Zhan , Bo Xu , Liang Yang , Hongfei Lin This is my paper · ORCID
classification cs.IR
keywords newsrecommendationentity-guidedinterestcontrastivepre-trainingdual-toweruserencodercrossattentionsignatureentityknowledge-graph-freeclickprediction
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

News recommendation usually treats named entities as extra semantic clues. This paper argues that entities play a distinct behavioral role: during the brief scanning of a headline one entity catches the eye, and during the later title-reading and click stage a reader's accumulated entity preferences guide whether the title is worth reading. The proposed IP2 probes the scanning-stage signal by compressing all entities in a headline into one signature entity vector and aligning it with the title through contrastive pretraining; it probes the second signal with a dual-tower user encoder where entity interest and title-meaning interest attend to each other before the click score is computed. On MIND-small, MIND-large, and Adressa-1week, IP2 reports the best AUC, MRR, nDCG@5, and nDCG@10 among all methods compared, including gains of about 0.8–1.0 AUC on MIND-small, 1.0–1.4 on MIND-large, and about 4.5 on Adressa. If the mechanism is real, entity guidance is not a side channel but a primary driver of reading and clicking behavior.

What carries the argument

The machinery that carries the argument is the signature entity encoder plus the dual-tower cross-attention user encoder, trained in two stages. The signature entity encoder is a small transformer over entity embeddings with a prepended [ent] token; during contrastive pretraining, the [ent] output is matched against the title embedding so the entity memory takes on meaning from the news text rather than from an external knowledge graph. The user encoder runs two parallel attention towers over the reading history, one querying titles against entities and the other querying entities against titles, and a learnable gate mixes the two user preference vectors; the same gating is applied to the candidate news embedding. The cross-attention link is the component that lets inter-news entity interest adjust title-reading interest, and the learnable aggregation is what the paper uses to adapt the strength of entity guidance per user and per news item.

What would settle it

An ablation that keeps the contrastive pretraining and dual-tower user encoder but replaces the signature entity encoder's [ent] output with the average of the encoded entity vectors would settle the mechanism: if the score drop is far below the roughly 1.0 AUC drop of the full w/o Intra ablation, the leading-entity probing is not the source of the reported gains.

Watch

Extended reading notes

Core claim

The paper's central claim is that a news recommender should model entity interest at exactly the two places the reading behavior chain puts it. At the intra-news level, a signature entity encoder prepends a learnable [ent] token to the entities mentioned in a title, stacks transformer layers over them, and uses the [ent] output as the title's signature entity; a contrastive pretraining loss pulls this signature entity toward the title embedding and pushes it away from other titles, which the paper interprets as teaching the entity memory which entity is the leading attractor during scanning. At the inter-news level, a dual-tower user encoder derives one preference vector from title meanings and one from entities, and cross-attention lets each tower query the other so that entity interest can calibrate title reading interest before the final click prediction. The paper reports that IP2 outperforms all compared neural and knowledge-aware baselines on every metric on MIND-small, MIND-large, and Adressa-1week, and that ablations removing the intra-news contrastive pretraining or the inter-news cross-attention each cost roughly 1.0 AUC on MIND-small, with the inter-news removal costing the most.

Load-bearing premise

The argument assumes that the entity encoder's summary vector really captures the single most attention-grabbing entity in a headline, rather than simply reproducing whatever the title means as a whole.

Editorial extensions

If this is right

  • If the two-level entity signal is the cause of the gains, news recommenders can build entity representations from unlabeled title corpora and click logs, avoiding knowledge graphs that age quickly and miss new entities.
  • The same scan-title-click decomposition gives a design template for other short-text recommendation domains, where a salient token often directs first-pass attention.
  • The entity tower and its gate weight provide a readable account of why an item was ranked: the gate says whether entity interest or title meaning dominated for that user.
  • Because the contrastive pretraining stage requires no interaction logs, the method can be warmed up on large news collections before seeing any user data, which the paper says helps on smaller datasets such as MIND-small.

Reading between the lines

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

  • One testable extension is to quantify how much of the gain comes from bridging unlabeled phrases: the case study suggests the cross-attention link connects a phrase like 'Black Friday' to a labeled entity even when the phrase itself is not annotated, so performance on headlines with very few annotated entities could be measured separately.
  • The behavioral story predicts measurable differences at the scanning stage itself, so an eye-tracking or dwell-time study on headlines with one dominant entity would test whether the [ent] representation tracks actual first-pass attention rather than just click patterns.
  • An ablation that replaces the signature entity encoder with a second copy of the title encoder, keeping the same contrastive and cross-attention setup, would separate the value of a dedicated entity stream from the value of having any second aligned stream at all.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The paper proposes IP2, a news recommendation model that aims to model entity-guided reading interest at two levels: an intra-news level, where a Signature Entity Encoder (SEE) aggregates entities in a title into a single 'signature entity' representation trained by contrastive pre-training against the title embedding, and an inter-news level, where a dual-tower user encoder uses cross attention between title and entity streams to combine semantic and entity-guided user interest. The model is evaluated on MIND-small, MIND-large, and Adressa-1week, with extensive baselines, ablations, and a case study, and the authors report state-of-the-art results. The paper also releases source code. The central SOTA claim is supported by the reported experiments, but the paper's behavioral interpretation of the intra-news mechanism is not directly evidenced, and there are reproducibility and statistical reporting issues.

Significance. If the empirical results are correct, IP2 is a practically useful contribution: it outperforms a broad set of neural and knowledge-aware baselines on three dataset variants, and it does so without requiring an external knowledge graph. The two-stage contrastive pre-training recipe and the cross-attention dual-tower user encoder are concrete, reproducible architectural ideas. The paper also provides a public code repository, which strengthens the reproducibility of the work. The main significance risk is that the paper frames its novelty around 'probing' the leading entity during scanning, but the loss function does not actually enforce that the signature entity corresponds to a leading entity or even to any single entity. The ablation results show that the entity-title contrastive loss is useful, but they do not establish the proposed cognitive interpretation. This needs to be addressed before the claimed mechanism can be accepted.

major comments (3)
  1. [Section 4.1.2 and Eq. (12)] The signature entity e_i is supervised only by the contrastive objective in Eq. (12), where the positive pair is (e_i, h_i) and h_i is the pooled embedding of the entire title. Nothing in this loss forces the [ent] output to concentrate on one leading entity, to match a particular entity embedding, or to be distinct from a generic title summary. Consequently, the claims in Section 4.2 and Figure 3 that SEE 'probes' intra-news entity interest and that the case-study attention weights in Section 5.6 reveal the leading entity are not established by the presented evidence. The 1.0 AUC drop of w/o Intra in Table 4 is equally consistent with e_i acting as a second, complementary view of the whole title. Please provide a direct test of the probing mechanism, for example by comparing SEE attention or e_i against human- or entity-level salience annotations, or by ablating with the [ent] output replaced by a mean-pooled entity representation.
  2. [Section 4.2 and Section 5.1.2] Section 4.2 states that the contrastive weights alpha, beta, and delta sum to 1, but Section 5.1.2 reports alpha=0.3, beta=0.2, delta=0.2, which sum to 0.7. The exact pre-training objective is therefore ambiguous and the reported setting cannot be reproduced as written. Please correct the inconsistency and state the exact loss weights used in the experiments.
  3. [Section 5.2, Tables 2 and 3] The headline improvements over the best baseline are modest, approximately 0.8 to 1.0 AUC on MIND-small and 1.0 to 1.4 AUC on MIND-large, yet only five-run averages are reported, without standard deviations, confidence intervals, or error bars. The significance statement that improvements are significant at the 0.05 level with a paired t-test does not say which baseline or baselines were tested. Please report per-seed variance and state the exact comparison protocol. In addition, Table 3 compares only four baselines on Adressa, so the claim that IP2 'outperforms all compared methods in all cases' should be qualified by noting that knowledge-graph-aware baselines could not be run on that dataset.
minor comments (3)
  1. [Section 5.4 and Figure 4] The caption and axis labels of Figure 4 are difficult to parse: 'Random' and 'TransE' appear on the x-axis, but the curves are labeled only by 'w/ CP' and 'w/o CP', and no numerical values or error bars are shown. Please add a clear legend and report numeric results with variance.
  2. [Section 5.6 and Figure 7] The case study is qualitative and the attention visualization is post-hoc; the text should explicitly state that the highlighted attention weights are not supervised and should not be read as direct evidence for the probing mechanism.
  3. [Section 5.4] The text contains a typo, 'TranE', which should be 'TransE'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: entity representations are learned from news titles, not from click labels; the [ent]-to-title alignment is an interpretive gap, not a self-referential derivation.

full rationale

No circularity found. IP2's central empirical claim is tested on held-out click labels from MIND-small, MIND-large, and Adressa-1week (Section 5.2), and the compared methods are external baselines. The entity embeddings are produced by the contrastive objective in Eq. (9)-(12), whose positive pair is (e_i, h_i) with h_i the BERT title embedding; these embeddings are learned from news titles, not from the test click labels or from the reported AUC figures, so no fitted parameter is renamed as a prediction. The downstream recommendation score is the standard NCE loss (Eq. 18) over click labels, making the SOTA claim an external empirical result rather than a restatement of the model's construction. The main weakness is interpretive: because the [ent] output is only supervised to match the whole title, the paper's claim that it isolates a 'leading entity' (Section 4.2, Figure 3, Section 5.6) is underdetermined; the planned user studies in Section 6 effectively concede this. That is a mechanistic-validity gap, not a circular derivation. Self-citations ([31], [41]-[43]) appear only in related-work, as a baseline, or as generic technical statements and are not load-bearing. The alpha+beta+delta inconsistency between Section 4.2 ('sum up to 1') and Section 5.1.2 (0.3+0.2+0.2=0.7) is a correctness/typo issue, not circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central claim rests on a behavioral model of news reading, on the completeness of dataset entity annotations, and on the assumption that contrastively aligned entity representations isolate leading-entity interest. None of these is directly validated in the paper. The free parameters (loss weights, temperature, number of layers, negatives) are chosen manually and tuned on MIND-small.

free parameters (5)
  • Contrastive loss weights alpha, beta, delta = 0.3, 0.2, 0.2
    Set manually for IP2; text claims they sum to 1 but they sum to 0.7. Tuned on MIND-small.
  • Temperature tau = 0.1
    Used in all contrastive losses; chosen by hand.
  • Number of SEE transformer layers L = 2
    Selected via sweep over {1,2,3,4,5} on MIND-small; performance peaks at L=2.
  • Number of negative samples r = 4
    Chosen for downstream NCE loss; standard in MIND-style evaluation.
  • Batch sizes and learning rates = pretrain batch 128, downstream batch 64; lr 1e-5 BERT, 1e-4 others
    Manually set hyperparameters; not shown to be robust across datasets.
assumptions (5)
  • domain assumption News reading consists of three successive steps: scanning, title reading, then clicking, with entity interest guiding each step.
    Motivates the architecture and the two-level entity interest design; based on a single behavioral reference (Liu 2005) and not tested in this paper.
  • domain assumption The entity annotations provided by MIND and Adressa are sufficient to identify the entities that matter; no NER is performed.
    The entity encoder and entity tower depend entirely on these annotations; the case study itself notes that 'Black Friday' is not labeled as an entity, showing the assumption can fail.
  • ad hoc to paper The [ent] output of the signature entity encoder, after contrastive pre-training, represents the leading entity and intra-news scanning interest.
    This is the core operationalization of intra-news interest; the paper infers it from downstream accuracy and one case study rather than demonstrating it directly (Section 4.2, Figure 3).
  • ad hoc to paper Cross attention between the title and entity towers models the behavioral influence of inter-news entity interest on title reading.
    Used to implement inter-news guidance; validated only via ablation (w/o Inter), not via behavioral data (Section 4.3.1).
  • domain assumption Pretrained language models encode sufficient entity knowledge for contrastive alignment to initialize useful entity embeddings.
    The pre-training stage relies on BERT to provide title semantics that the entity representations are aligned to; the paper argues PLMs can describe entities from multiple angles (Section 4.2).
invented entities (2)
  • Signature entity representation
    purpose: A single embedding summarizing all entities in a news title, produced by the [ent] token of a Transformer over entity embeddings.
    A learned latent construct with no external ground truth; the paper only offers downstream performance and a case study as evidence.
  • Handle token [ent]
    purpose: A prepended token analogous to BERT's [CLS], used to aggregate entity information into one vector.
    An architectural invention; its output is assumed to be meaningful but is not externally validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IP2: Entity-Guided Interest Probing for Personalized News Recommendation." pith.science (2026). https://pith.science/paper/VMVILIC7

@misc{pith2026250713622,
  author       = {Pith},
  title        = {Pith review of: IP2: Entity-Guided Interest Probing for Personalized News Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VMVILIC7}},
  note         = {Machine review of arXiv:2507.13622}
}
read the original abstract

News recommender systems aim to provide personalized news reading experiences for users based on their reading history. Behavioral science studies suggest that screen-based news reading contains three successive steps: scanning, title reading, and then clicking. Adhering to these steps, we find that intra-news entity interest dominates the scanning stage, while the inter-news entity interest guides title reading and influences click decisions. Unfortunately, current methods overlook the unique utility of entities in news recommendation. To this end, we propose a novel method called IP2 to probe entity-guided reading interest at both intra- and inter-news levels. At the intra-news level, a Transformer-based entity encoder is devised to aggregate mentioned entities in the news title into one signature entity. Then, a signature entity-title contrastive pre-training is adopted to initialize entities with proper meanings using the news story context, which in the meantime facilitates us to probe for intra-news entity interest. As for the inter-news level, a dual tower user encoder is presented to capture inter-news reading interest from both the title meaning and entity sides. In addition to highlighting the contribution of inter-news entity guidance, a cross-tower attention link is adopted to calibrate title reading interest using inter-news entity interest, thus further aligning with real-world behavior. Extensive experiments on two real-world datasets demonstrate that our IP2 achieves state-of-the-art performance in news recommendation.

Figures

Figures reproduced from arXiv: 2507.13622 by the authors.

Figure 1
Figure 1. An example of a complete news selection chain in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of IP2. IP2 follows a two-stage training paradigm. In the first stage, we conduct signature entity-title [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The architecture of Signature Entity Encoder. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Results on different SEE entity memory setups. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Impact of the BERT size. 1 2 3 4 5 L 68.75 69.00 69.25 69.50 69.75 AUC 1 2 3 4 5 L 33.50 33.75 34.00 34.25 34.50 MRR 1 2 3 4 5 L 37.25 37.50 37.75 38.00 38.25 nDCG@5 1 2 3 4 5 L 43.50 43.75 44.00 44.25 44.50 nDCG@10 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Impact of the SEE size. 5.6 Case Study (RQ5) We further conduct a case study to illustrate IP2’s effectiveness in real-world cases. As shown in [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Case study based on a sampled impression log. Enti [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 40 canonical work pages

  1. [1]

    Mingxiao An, Fangzhao Wu, Chuhan Wu, Kun Zhang, Zheng Liu, and Xing Xie

    work page

  2. [2]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. 2016. Layer normaliza- tion. arXiv preprint arXiv:1607.06450 (2016)

    arXiv 2016

  3. [3]

    Bao Chen, Yong Xu, Jingru Zhen, Xin He, Qun Fang, and Jinde Cao. 2024. NRMG: News Recommendation With Multiview Graph Convolutional Networks. IEEE Transactions on Computational Social Systems 11 (2024), 2245–2255

    work page 2024

  4. [4]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A Simple Framework for Contrastive Learning of Visual Representations. In ICML (Proceedings of Machine Learning Research, Vol. 119) , Hal Daumé III and Aarti Singh (Eds.). PMLR, 1597–1607

    work page 2020

  5. [5]

    Chong Feng, Muzammil Khan, Arif Ur Rahman, and Arshad Ahmad. 2020. News recommendation systems-accomplishments, challenges & future directions. IEEE Access 8 (2020), 16702–16725

    work page 2020

  6. [6]

    Natalie Fenton. 2009. News in the digital age. In The Routledge companion to news and journalism. Routledge, 557–567

    work page 2009

  7. [7]

    Suyu Ge, Chuhan Wu, Fangzhao Wu, Tao Qi, and Yongfeng Huang. 2020. Graph enhanced representation learning for news recommendation. In Proceedings of the web conference 2020 . 2863–2869

    work page 2020

  8. [8]

    Jon Atle Gulla, Lemei Zhang, Peng Liu, Özlem Özgöbek, and Xiaomeng Su. 2017. The adressa dataset for news recommendation. In Proceedings of the international conference on web intelligence . 1042–1048

    work page 2017

Show all 46 references
  1. [9]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction. In IJCAI. 1725–1731

    2017

  2. [10]

    Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In Proceedings of the 26th international conference on world wide web . 173–182

    2017

  3. [11]

    Hiroshi Inoue. 2019. Multi-sample dropout for accelerated training and better generalization. arXiv preprint arXiv:1905.09788 (2019)

    2019 arXiv

  4. [12]

    Jia Hua Jeng. 2024. Bridging Viewpoints in News with Recommender Systems. In Proceedings of the 18th ACM Conference on Recommender Systems . 1283–1289

    2024

  5. [13]

    Nithish Kannen, Yao Ma, Gerrit Van Den Burg, and Jean Baptiste Faddoul. 2024. Efficient Pointwise-Pairwise Learning-to-Rank for News Recommendation. In Findings of the Association for Computational Linguistics: EMNLP 2024 . 12403– 12418

    2024

  6. [14]

    Jacob Devlin Ming-Wei Chang Kenton and Lee Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In NAACL-HLT. 4171–4186

    2019

  7. [15]

    Julien Leblay and Melisachew Wudage Chekol. 2018. Deriving validity time in knowledge graph. In Companion Proceedings of the The Web Conference 2018 . 1771–1776

    2018

  8. [16]

    Miaomiao Li and Licheng Wang. 2019. A survey on personalized news recom- mendation technology. IEEE Access 7 (2019), 145861–145879

    2019

  9. [17]

    Danyang Liu, Jianxun Lian, Shiyin Wang, Ying Qiao, Jiun-Hung Chen, Guangzhong Sun, and Xing Xie. 2020. KRED: Knowledge-aware document repre- sentation for news recommendations. In Proceedings of the 14th ACM conference on recommender systems. 200–209

    2020

  10. [18]

    Rui Liu, Bin Yin, Ziyi Cao, Qianchen Xia, Yong Chen, and Dell Zhang. 2023. Per- CoNet: News Recommendation with Explicit Persona and Contrastive Learning. arXiv preprint arXiv:2304.07923 (2023)

    2023 arXiv

  11. [19]

    Ziming Liu. 2005. Reading behavior in the digital environment: Changes in reading behavior over the past ten years. Journal of documentation 61, 6 (2005), 700–712

    2005

  12. [20]

    Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. 2019. Vilbert: Pretrain- ing task-agnostic visiolinguistic representations for vision-and-language tasks. Advances in neural information processing systems 32 (2019)

    2019

  13. [21]

    Guangyuan Ma, Hongtao Liu, W Xing, Wanhui Qian, Zhepeng Lv, Qing Yang, and Songlin Hu. 2023. PUNR: Pre-training with User Behavior Modeling for News Recommendation. In Findings of the Association for Computational Linguistics: EMNLP 2023. 8338–8347

    2023

  14. [22]

    Zhiming Mao, Jian Li, Hongru Wang, Xingshan Zeng, and Kam-Fai Wong. 2022. DIGAT: Modeling News Recommendation with Dual-Graph Interaction. In Find- ings of the Association for Computational Linguistics: EMNLP 2022 . 6595–6607

    2022

  15. [23]

    Sosuke Nishikawa, Ryokan Ri, Ikuya Yamada, Yoshimasa Tsuruoka, and Isao Echizen. 2022. EASE: Entity-Aware Contrastive Learning of Sentence Embedding. In NAACL-HLT. 3870–3885

    2022

  16. [24]

    Jay Pujara, Eriq Augustine, and Lise Getoor. 2017. Sparsity and noise: Where knowledge graph embeddings fall short. In Proceedings of the 2017 conference on empirical methods in natural language processing . 1751–1756

    2017

  17. [25]

    Tao Qi, Fangzhao Wu, Chuhan Wu, and Yongfeng Huang. 2021. Personalized news recommendation with knowledge-aware interactive matching. In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 61–70

    2021

  18. [26]

    Tao Qi, Fangzhao Wu, Chuhan Wu, and Yongfeng Huang. 2022. FUM: Fine- grained and Fast User Modeling for News Recommendation. Proceedings of the 45th international ACM SIGIR conference on research and development in information retrieval (2022)

    2022

  19. [27]

    Tao Qi, Fangzhao Wu, Chuhan Wu, Peiru Yang, Yang Yu, Xing Xie, and Yongfeng Huang. 2021. HieRec: Hierarchical User Interest Modeling for Personalized News Recommendation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th In...

    2021

  20. [28]

    Zhaopeng Qiu, Yunfan Hu, and Xian Wu. 2022. Graph neural news recommen- dation with user existing and potential interest modeling. ACM Transactions on Knowledge Discovery from Data (TKDD) 16, 5 (2022), 1–17

    2022

  21. [29]

    Steffen Rendle. 2012. Factorization machines with libfm. ACM Transactions on Intelligent Systems and Technology (TIST) 3, 3 (2012), 1–22

    2012

  22. [30]

    Steffen Rendle, Walid Krichene, Li Zhang, and John Anderson. 2020. Neural collaborative filtering vs. matrix factorization revisited. In Proceedings of the 14th ACM Conference on Recommender Systems . 240–248

    2020

  23. [31]

    Yijie Shu, Xiaokun Zhang, Youlin Wu, Bo Xu, Liang Yang, and Hongfei Lin

  24. [32]

    Feng Wang and Huaping Liu. 2021. Understanding the behaviour of contrastive loss. In CVPR. 2495–2504

    2021

  25. [33]

    Hongwei Wang, Fuzheng Zhang, Xing Xie, and Minyi Guo. 2018. DKN: Deep knowledge-aware network for news recommendation. In Proceedings of the 2018 world wide web conference . 1835–1844

    2018

  26. [34]

    Chuhan Wu, Fangzhao Wu, Suyu Ge, Tao Qi, Yongfeng Huang, and Xing Xie. 2019. Neural news recommendation with multi-head self-attention. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natur...

    2019

  27. [35]

    Chuhan Wu, Fangzhao Wu, Yongfeng Huang, and Xing Xie. 2021. User-as-Graph: User Modeling with Heterogeneous Graph Pooling for News Recommendation.. In IJCAI. 1624–1630

    2021

  28. [36]

    Chuhan Wu, Fangzhao Wu, Tao Qi, and Yongfeng Huang. 2021. Empowering news recommendation with pre-trained language models. In Proceedings of the 44th international ACM SIGIR conference on research and development in informa- tion retrieval. 1652–1656

    2021

  29. [37]

    Fangzhao Wu, Ying Qiao, Jiun-Hung Chen, Chuhan Wu, Tao Qi, Jianxun Lian, Danyang Liu, Xing Xie, Jianfeng Gao, Winnie Wu, and Ming Zhou. 2020. MIND: A Large-scale Dataset for News Recommendation. In ACL. Association for Com- putational Linguistics, Online, 3597–3606

    2020

  30. [38]

    Xiongfeng Xiao, Qing Li, Songlin Liu, and Kun Zhou. 2023. Improving News Recommendation via Bottlenecked Multi-task Pre-training. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2082–2086

    2023

  31. [39]

    Boming Yang, Dairui Liu, Toyotaro Suzumura, Ruihai Dong, and Irene Li. 2023. Going Beyond Local: Global Graph-Enhanced Personalized News Recommenda- tions. In Proceedings of the 17th ACM conference on recommender systems . 24–34

    2023

  32. [40]

    Qi Zhang, Jingjie Li, Qinglin Jia, Chuyuan Wang, Jieming Zhu, Zhaowei Wang, and Xiuqiang He. 2021. UNBERT: User-News Matching BERT for News Recom- mendation.. In IJCAI, Vol. 21. 3356–3362

    2021

  33. [41]

    Xiaokun Zhang, Bo Xu, Chenliang Li, Bowei He, Hongfei Lin, Chen Ma, and Fenglong Ma. 2025. A Survey on Side Information-driven Session-based Recom- mendation: From a Data-centric Perspective. IEEE Transactions on Knowledge and Data Engineering (2025)

    2025

  34. [42]

    Xiaokun Zhang, Bo Xu, Fenglong Ma, Zhizheng Wang, Liang Yang, and Hongfei Lin. 2025. Rethinking contrastive learning in session-based recommendation. Pattern Recognition (2025), 111924

    2025

  35. [43]

    Xiaokun Zhang, Bo Xu, Youlin Wu, Yuan Zhong, Hongfei Lin, and Fenglong Ma

  36. [46]

    In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval

    Finerec: Exploring fine-grained sequential recommendation. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1599–1608

  37. [2019]

    Neural news recommendation with long-and short-term user representa- tions. In ACL. 336–345

  38. [2024]

    In CCF International Conference on Natural Language Processing and Chinese Computing

    Don’t Click the Bait: Title Debiasing News Recommendation via Cross- Field Contrastive Learning. In CCF International Conference on Natural Language Processing and Chinese Computing . Springer, 224–236

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.