Pith. sign in

REVIEW 3 major objections 5 minor 41 references

Towards Multi-Aspect Diversification of News Recommendations Using Neuro-Symbolic AI for Individual and Societal Benefit

T0 review · 3 major / 5 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 The paper argues that news recommendation diversity should be treated as a multi-aspect, four-mode problem and proposes a neuro-symbolic approach—knowledge graphs plus learned rules—to handle it.

desk verdict A useful, honest idea paper with a solid four-mode taxonomy for multi-aspect news diversity, but the planned evaluation can't back the depolarization claim it reaches for. read the letter →

arxiv 2509.02220 v1 pith:XX5CHW52 submitted 2025-09-02 cs.IR cs.AI

Markus Reiter-Haas , Elisabeth Lex This is my paper
classification cs.IRcs.AI
keywords newsrecommendersystemsmulti-aspectdiversityframingknowledgegraphsrulelearningneuro-symbolicAIbalancedconsumptionevaluation
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 idea paper contends that news recommenders currently diversify at most one aspect at a time—typically topic or viewpoint—and that this is not enough. It defines multi-aspect diversity as the average pairwise distance between recommended items over several aspects at once (topic and framing in the guiding example) and extends the challenge to four recommendation modes: lists, sequences, LLM-generated summaries, and user interactions. The research direction it advocates is neuro-symbolic: knowledge graphs structure the aspects and label similarities, diversity-aware training objectives make models optimize diversity end-to-end, and transparent learned rules constrain and explain the system. The payoff, if the direction works, is a more balanced news diet for individuals, more serendipity, and less polarization for society.

What carries the argument

The carrying identity is the multi-aspect diversity measure div(D) = (1/(|D|(|D|-1))) sum over distinct pairs of dist(d_i,d_j), where dist = alpha * dist_topic + (1-alpha) * dist_frame and the set of aspects can be extended beyond two. The alpha parameter allows trading aspects against each other, and the pairwise average makes the measure order-independent within a list. This identity drives the entire argument: it makes the insufficiency of single-aspect diversity concrete (0.33 vs 0.75 in the example), defines a common objective for all four modes, and is what the knowledge-graph and rule-learning machinery is meant to optimize.

What would settle it

Give the same set of articles two different annotations: one with a single topic and frame per article, one with multiple labels and their weights; if the two diversity scores disagree strongly on real corpora, the single-label formalization fails. Alternatively, run a randomized user study comparing a topic-only diversified newsletter against a topic-and-frame diversified newsletter; if users do not perceive or benefit from the two-aspect version, the claimed value of multi-aspect diversification is not supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that diversity in news recommendations is a multi-aspect phenomenon that cannot be handled by a single metric applied to one mode. It formalizes the diversity of a collection as the average symmetric pairwise distance between documents, where each document carries labels for several aspects—topic and framing in the illustration—and the distance is a weighted combination of per-aspect distances with a tunable weight alpha. Working through a two-topic, four-frame example, it shows that diversifying only topics yields a low score (0.33), diversifying only framing yields 0.42, and only a list that balances both reaches 0.75. That example motivates the paper's program

Load-bearing premise

The load-bearing premise is that each news item can be summarized by one label per aspect (one topic and one frame) and that diversity is the average pairwise distance over those labels; the paper itself calls this 'rather unrealistic,' and if real news is multi-label, hierarchical, or context-dependent, the measure doesn't capture the intended diversity.

Editorial extensions

If this is right

  • A shared multi-aspect distance lets the same diversity objective be applied to lists, sequences, summaries, and interactions, so systems can be compared on one scale instead of mode-specific heuristics.
  • Optimizing diversity inside the model's loss, with a knowledge graph as regularizer, turns diversification from a post-processing swap into an end-to-end property.
  • Transparent rules over the knowledge graph make it possible to enforce global constraints (e.g., avoid misinformation), regional or contextual conditions, user-controlled filtering, and learned rules that can double as explanations.
  • User studies rather than offline accuracy are the evaluation plan, because the ideal diversity level is user-dependent and maximizing diversity can backfire if it forces opposing viewpoints on users.
  • Diversifying the framing of controversial topics, such as presenting immigration through a cultural angle for one group and a security angle for another, is proposed as a bridging mechanism that could moderate extreme positions.

Reading between the lines

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

  • Inference: the four-mode decomposition implies a comparative experiment the paper doesn't run—feed the same user the same articles as a list, a sequence, a summary, and an interactive feed, and check whether the mode-specific diversity metrics predict perceived diversity equally; if not, the distance function needs a mode-dependent or order-dependent term.
  • Inference: the single-topic/single-frame assumption can be stress-tested by annotating articles with multiple labels and comparing the paper's pairwise distance against a distribution-based one; multi-label articles would be systematically undercounted as diverse.
  • Inference: the paper's vision of a personal behavior score (rewards for engaging with opposing viewpoints) suggests a testable extension where interaction diversity is tracked per user and users are shown their own diversity score—if such feedback changes consumption patterns, diversification could become a user-facing feature rather than only a system-side optimization.
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

3 major / 5 minor

Summary. This idea paper argues that news recommender diversity should be treated as a multi-aspect problem, going beyond single aspects such as topic or viewpoint. It formalizes diversity as the average pairwise distance over document labels (Section 2), identifies four recommendation modes—lists, sequences, summaries, and interactions (Section 3)—and proposes a neuro-symbolic research direction combining knowledge graphs, learned rules, and mode-specific algorithms (Section 4.1). The evaluation plan (Section 4.2) is based on offline datasets plus online user studies, with POPROX described as an example platform. The paper's stated vision is that multi-aspect diversification yields individual benefits (e.g., serendipity) and societal benefits (e.g., reduced polarization), but it reports no experiments and explicitly frames the work as a research direction.

Significance. If realized, the proposed framework would integrate a fragmented literature on news diversity and extend it to timely novel modes such as LLM-generated summaries and user-interaction diversity. The paper is well structured, gives a clear taxonomy of four modes, and is honest about many of its simplifying assumptions. Its main strength is the problem formulation and the identification of open challenges, not a validated method. The societal and individual benefit claims are hypotheses, and—as I detail below—the proposed evaluation is not designed to test the societal half of those claims. This gap is central to the paper's stated purpose.

major comments (3)
  1. [§4.2 and §4.3] The paper's central claim is that multi-aspect diversification provides both individual and societal benefit, including decreased polarization and balanced news consumption (Abstract; §4.3). However, the planned POPROX randomized controlled trial measures only perceived experience—enjoyment, engagement, news fatigue, serendipity, positivity—via weekly surveys. There is no outcome variable for political polarization, attitude change, or consumption balance. Since the paper itself cites Bail et al. [35] showing that exposure to opposing views can increase polarization, the asserted positive societal effect is not a foregone conclusion. The evaluation plan is therefore structurally unable to support the societal-benefit half of the claim. The authors should either add pre/post measures of polarization or balanced-consumption outcomes, or explicitly re-scope the paper's claim to individual p
  2. [§2, diversity formalization and distance function] The formalization assumes each document has exactly one topic label and one frame label, and defines diversity as the average symmetric pairwise distance over these labels. The paper acknowledges the one-label assumption is 'rather unrealistic,' but the issue is more than cosmetic: all four modes in §3 and the 'generalized metrics' in §4.1 inherit this representation. Furthermore, the distance values and the α weighting are chosen by hand for the illustration, with no indication of how they would be learned, calibrated, or validated for real news corpora. For an idea paper this is acceptable, but the research direction would be substantially stronger if it included at least a parameterized family of metrics with a discussion of multi-label, hierarchical, or context-dependent aspects—or an explicit statement that these are open problems, rather than a single fixed definition.
  3. [§4.1, algorithmic proposals] The proposed neuro-symbolic components are described only as plans: 'we can incorporate knowledge graphs,' 'we plan to incorporate transparent rules,' 'learned rules could lead to increased transparency.' There is no specification of the learning objective, the rule language, the interaction between symbolic rules and subsymbolic models, or the computational complexity. As a vision paper, this is a legitimate mode of contribution, but the title's 'Using Neuro-Symbolic AI' is stronger than what the manuscript actually presents. The authors should either provide a minimal proof-of-concept of at least one component (e.g., a rule operating on the §2 example) or clearly state that no implementation exists and that the contribution is exclusively a problem formulation and research agenda.
minor comments (5)
  1. [§1, references [12] and [13]] Both Wu et al. works are cited closely together; the attributions should be disambiguated so that the reader knows which 'both modes' and 'end-to-end diversity-aware' claims each reference supports.
  2. [Figure 1] The diversity values in Figure 1 (0, 0.33, 0.42, 0.75) are not derived in the text. A short calculation for one of the subfigures would help readers verify the metric.
  3. [§3.2] The claim 'recommending the item that is more dissimilar to recent items is preferable' would benefit from a precise definition of the 'recent' window and how it interacts with the list-level diversity metric.
  4. [§4.2] The sentence 'aiming for the expected (i.e., mean) diversity per user would not lead to an increase thereof' is confusing as written; clarify that targeting the observed mean is not the same as setting a higher target diversity level.
  5. [References] Reference [39] is a self-citation used as motivation; the connection between the 'framing loop' and the proposed diversification approach could be made more explicit in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an idea/position paper with no fitted predictions; its formalization is explicitly illustrative and its only self-citation is motivational.

full rationale

The paper does not derive an empirical result from fitted data. Section 2 introduces a diversity measure as average pairwise distance over hand-assigned topic and frame labels, but it explicitly labels this as illustrative: 'For illustration, we define two topics T = {Climate, Immigration} and four frames... and set α = 0.5.' These values are not estimated from data and no prediction is made from them, so there is no fitted-input-called-prediction pattern. The formalization's own acknowledged limitation—'this assumption is rather unrealistic'—is stated by the authors and is a modeling simplification, not a circular step. The only self-citation is [39] in Section 4.3, used to motivate why diversified framing matters ('news consumer tend to repeatedly consume similarly framed news [39]'); this is background motivation, not a load-bearing premise from which the proposed approach is derived, and it does not by itself force the paper's conclusions. The planned user studies in Section 4.2 are future work and do not yet produce predictions; one could question whether the weekly surveys fully operationalize the societal-benefit claim (they measure enjoyment, engagement, fatigue, serendipity, and positivity, but not polarization directly), but that is an evaluation-design/validity concern, not circularity. The paper's central contribution is a proposed research direction combining knowledge graphs, learned rules, and four recommendation modes, and that proposal is independent of any fitted parameter or self-citation chain. Therefore, the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or formal entities. Its free parameters are illustrative hand-chosen values in the toy example. The axioms are mostly domain assumptions about labelability, extractability, and the positive effects of diversity, all acknowledged or planned for evaluation.

free parameters (2)
  • Aspect weighting alpha = 0.5 (illustrative)
    Hand-set in Section 2 to weight topic versus framing distance in the diversity measure; a real system would need to set or learn this value.
  • Pairwise framing distances = e.g., Health-Cultural = 0.5, Health-Security = 1.0
    Hand-assigned in Section 2 to create the illustrative four-document example; not learned from data or justified by external evidence.
assumptions (4)
  • standard math The distance function is symmetric and satisfies dist(di, di) = 0.
    Assumed in Section 2 before defining div(D); this is a standard metric-like condition but is not empirically validated for document diversity.
  • domain assumption Each document can be represented by exactly one topic label and one frame label.
    Stated in Section 2: 'for simplicity, we assume each document di consists of only a topic label ti and frame label fi'. This ignores multi-label, hierarchical, and context-dependent aspects that the paper later acknowledges as realistic.
  • domain assumption Aspects can be extracted from content via topic and framing classifiers and represented as knowledge graphs.
    Section 4.1 states 'These aspects can be modeled as a knowledge graph ... which can be extracted with a topic and framing classifier from the content.' No implementation or accuracy evidence is provided.
  • domain assumption Increasing diversity in news consumption yields beneficial individual and societal effects, including serendipity and reduced polarization.
    Sections 4.2 and 4.3 present this as the research vision and cite supporting literature. It remains a hypothesis to be tested in the planned user studies, not an established fact.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Multi-Aspect Diversification of News Recommendations Using Neuro-Symbolic AI for Individual and Societal Benefit." pith.science (2026). https://pith.science/paper/XX5CHW52

@misc{pith2026250902220,
  author       = {Pith},
  title        = {Pith review of: Towards Multi-Aspect Diversification of News Recommendations Using Neuro-Symbolic AI for Individual and Societal Benefit},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XX5CHW52}},
  note         = {Machine review of arXiv:2509.02220}
}
read the original abstract

News recommendations are complex, with diversity playing a vital role. So far, existing literature predominantly focuses on specific aspects of news diversity, such as viewpoints. In this paper, we introduce multi-aspect diversification in four distinct recommendation modes and outline the nuanced challenges in diversifying lists, sequences, summaries, and interactions. Our proposed research direction combines symbolic and subsymbolic artificial intelligence, leveraging both knowledge graphs and rule learning. We plan to evaluate our models using user studies to not only capture behavior but also their perceived experience. Our vision to balance news consumption points to other positive effects for users (e.g., increased serendipity) and society (e.g., decreased polarization).

Figures

Figures reproduced from arXiv: 2509.02220 by the authors.

Figure 2
Figure 2. Diversifying news recommendations in different modes. (a) Lists can be diversified by swapping articles with others that have different aspects present. (b) Sequences need to consider items in a specific window, where the focus lies more on the similarity towards more recent items. Arrows denote similarity to previous items. (c) Summaries can be diversified at content or source-level. (d) Considering diversified int… view at source ↗
Figure 3
Figure 3. Integrating symbolic knowledge into news recommendations. Different aspects can be modeled as a knowledge graph (left) on which rules can operate (right). user behavior. 1. Globally defined rules are hardcoded in the system, e.g., to avoid spreading potential misinformation. Global rules are important for recommender platforms to adhere to regulations and internal goals. 2. Context-specific rules can be set to affec… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 39 canonical work pages

  1. [35]

    C. A. Bail, L. P. Argyle, T. W. Brown, J. P. Bumpus, H. Chen, M. F. Hunzaker, J. Lee, M. Mann, F. Merhout, A. Volfovsky, Exposure to opposing views on social media can increase political polarization, Proceedings of the National Academy of Sciences 115 (2018) 9216–9221

    work page 2018

  2. [1]

    Karimi, D

    M. Karimi, D. Jannach, M. Jugovac, News recommender systems–survey and roads ahead, Infor- mation Processing & Management 54 (2018) 1203–1227

    work page 2018

  3. [2]

    Diversity in News Recommendations

    A. Bernstein, C. De Vreese, N. Helberger, W. Schulz, K. Zweig, C. Baden, M. A. Beam, M. P. Hauer, L. Heitz, P. Jürgens, et al., Diversity in news recommendations, arXiv preprint arXiv:2005.09495 abs/2005.09495 (2020)

    work page Pith review arXiv 2005

  4. [3]

    Kulshrestha, M

    J. Kulshrestha, M. Zafar, L. Noboa, K. Gummadi, S. Ghosh, Characterizing information diets of social media users, in: Proceedings of the international AAAI conference on web and social media, volume 9, AAAI Press, 2015, pp. 218–227

    work page 2015

  5. [4]

    S. Raza, C. Ding, News recommender system: a review of recent progress, challenges, and opportunities, Artificial Intelligence Review 55 (2022) 1–52

    work page 2022

  6. [5]

    Kühne, C

    R. Kühne, C. Schemer, The emotional effects of news frames on information processing and opinion formation, Communication Research 42 (2015) 387–407

    work page 2015

  7. [6]

    Helberger, On the democratic role of news recommenders, in: Algorithms, automation, and news, Routledge, 2021, pp

    N. Helberger, On the democratic role of news recommenders, in: Algorithms, automation, and news, Routledge, 2021, pp. 14–33

    2021

  8. [7]

    Vrijenhoek, M

    S. Vrijenhoek, M. Kaya, N. Metoui, J. Möller, D. Odijk, N. Helberger, Recommenders with a mission: assessing diversity in news recommendations, in: Proceedings of the 2021 conference on human information interaction and retrieval, ACM, 2021, pp. 173–183

    work page 2021

Show all 41 references
  1. [8]

    Draws, N

    T. Draws, N. Tintarev, U. Gadiraju, Assessing viewpoint diversity in search results using ranking fairness metrics, ACM SIGKDD Explorations Newsletter 23 (2021) 50–58

    2021

  2. [9]

    Mulder, O

    M. Mulder, O. Inel, J. Oosterman, N. Tintarev, Operationalizing framing to support multiperspective recommendations of opinion pieces, in: Proceedings of the 2021 ACM conference on fairness, accountability, and transparency, ACM, 2021, pp. 478–488

    2021

  3. [10]

    Loecherbach, J

    F. Loecherbach, J. Moeller, D. Trilling, W. van Atteveldt, The unified framework of media diversity: A systematic literature review, Digital Journalism 8 (2020) 605–642

    2020

  4. [11]

    Mattis, P

    N. Mattis, P. Masur, J. Möller, W. van Atteveldt, Nudging towards news diversity: A theoretical framework for facilitating diverse news consumption through recommender design, New Media & Society 26 (2024) 3681–3706

    2024

  5. [12]

    C. Wu, F. Wu, Y. Huang, X. Xie, Personalized news recommendation: Methods and challenges, ACM Transactions on Information Systems 41 (2023) 1–50

    2023

  6. [13]

    C. Wu, F. Wu, T. Qi, Y. Huang, End-to-end learnable diversity-aware news recommendation, arXiv preprint arXiv:2204.00539 (2022)

    2022 arXiv

  7. [14]

    M. Kaya, D. G. Bridge, Accurate and diverse recommendations using item-based subprofiles., in: FLAIRS, 2018, pp. 462–467

    2018

  8. [15]

    J. Zhou, E. Agichtein, S. Kallumadi, Diversifying multi-aspect search results using simpson’s diversity index, in: Proceedings of the 29th ACM International conference on information & knowledge management, 2020, pp. 2345–2348

    2020

  9. [16]

    R. Li, L. Heitz, O. Inel, A. Bernstein, D-rdw: Diversity-driven random walks for news recommender systems, arXiv preprint arXiv:2508.13035 (2025)

    2025 arXiv

  10. [17]

    Vercoutere, G

    S. Vercoutere, G. Joris, T. De Pessemier, L. Martens, Improving selection diversity using hybrid graph-based news recommenders, User Modeling and User-Adapted Interaction 34 (2024) 955–993

    2024

  11. [18]

    A. Iana, G. Glavaš, H. Paulheim, Train once, use flexibly: A modular framework for multi-aspect neural news recommendation, in: Findings of the Association for Computational Linguistics: EMNLP 2024, 2024, pp. 9555–9571

    2024

  12. [19]

    Kunaver, T

    M. Kunaver, T. Požrl, Diversity in recommender systems–a survey, Knowledge-based systems 123 (2017) 154–162

    2017

  13. [20]

    Carraro, et al., Neuro-Symbolic Recommender Systems, Ph.D

    T. Carraro, et al., Neuro-Symbolic Recommender Systems, Ph.D. thesis, Università degli Studi di Padova, 2025

    2025

  14. [21]

    Ziegler, S

    C.-N. Ziegler, S. M. McNee, J. A. Konstan, G. Lausen, Improving recommendation lists through topic diversification, in: Proceedings of the 14th international conference on World Wide Web, ACM, 2005, pp. 22–32

    2005

  15. [22]

    X. Li, G. Cong, G. Xiao, Y. Xu, W. Jiang, K. Li, On evaluation metrics for diversity-enhanced recommendations, in: Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, ACM, 2024, pp. 1286–1295

    2024

  16. [23]

    H. Wu, Y. Zhang, C. Ma, F. Lyu, B. He, B. Mitra, X. Liu, Result diversification in search and recommendation: A survey, IEEE Transactions on Knowledge and Data Engineering 36 (2024) 5354–5373

    2024

  17. [24]

    A. L. Opdahl, T. Al-Moslmi, D.-T. Dang-Nguyen, M. Gallofré Ocaña, B. Tessem, C. Veres, Semantic knowledge graphs for the news: A review, ACM Computing Surveys 55 (2022) 1–38

    2022

  18. [25]

    Q. Guo, F. Zhuang, C. Qin, H. Zhu, X. Xie, H. Xiong, Q. He, A survey on knowledge graph- based recommender systems, IEEE Transactions on Knowledge and Data Engineering 34 (2020) 3549–3568

    2020

  19. [26]

    S. A. Puthiya Parambath, N. Usunier, Y. Grandvalet, A coverage-based approach to recommendation diversity on similarity graph, in: Proceedings of the 10th ACM Conference on Recommender Systems, ACM, 2016, pp. 15–22

    2016

  20. [27]

    R. Xie, Q. Liu, S. Liu, Z. Zhang, P. Cui, B. Zhang, L. Lin, Improving accuracy and diversity in matching of recommendation with diversified preference network, IEEE Transactions on Big Data 8 (2021) 955–967

    2021

  21. [28]

    D. Liu, T. Bai, J. Lian, X. Zhao, G. Sun, J.-R. Wen, X. Xie, News graph: An enhanced knowledge graph for news recommendation., KaRS@ CIKM 1 (2019)

    2019

  22. [29]

    M. Beck, K. Pöppel, M. Spanring, A. Auer, O. Prudnikova, M. Kopp, G. Klambauer, J. Brandstetter, S. Hochreiter, xlstm: Extended long short-term memory, Advances in Neural Information Processing Systems 37 (2024) 107547–107603

    2024

  23. [30]

    Manoharan, R

    S. Manoharan, R. Senthilkumar, An intelligent fuzzy rule-based personalized news recommendation using social media mining, Computational intelligence and neuroscience 2020 (2020) 3791541

    2020

  24. [31]

    W. Ma, M. Zhang, Y. Cao, W. Jin, C. Wang, Y. Liu, S. Ma, X. Ren, Jointly learning explainable rules for recommendation with knowledge graph, in: The world wide web conference, ACM, 2019, pp. 1210–1221

    2019

  25. [32]

    Kruse, K

    J. Kruse, K. Lindskow, S. Kalloori, M. Polignano, C. Pomo, A. Srivastava, A. Uppal, M. R. Andersen, J. Frellsen, Eb-nerd a large-scale dataset for news recommendation, in: Proceedings of the Recommender Systems Challenge 2024, ACM, 2024, pp. 1–11

    2024

  26. [33]

    F. Wu, Y. Qiao, J.-H. Chen, C. Wu, T. Qi, J. Lian, D. Liu, X. Xie, J. Gao, W. Wu, et al., Mind: A large-scale dataset for news recommendation, in: Proceedings of the 58th annual meeting of the association for computational linguistics, Association for Computational Linguistics...

    2020

  27. [34]

    J. A. Gulla, L. Zhang, P. Liu, Ö. Özgöbek, X. Su, The adressa dataset for news recommendation, in: Proceedings of the international conference on web intelligence, ACM, 2017, pp. 1042–1048

    2017

  28. [36]

    W. Wu, L. Chen, Y. Zhao, Personalizing recommendation diversity based on user personality, User Modeling and User-Adapted Interaction 28 (2018) 237–276

    2018

  29. [37]

    Burke, M

    R. Burke, M. Ekstrand, Conducting recommender systems user studies using poprox, in: Adjunct Proceedings of the 33rd ACM Conference on User Modeling, Adaptation and Personalization, ACM, 2025, pp. 1–2

    2025

  30. [38]

    C. A. Bail, D. S. Hillygus, A. Volfovsky, M. Allamong, F. Alqabandi, D. M. Jordan, G. Tierney, C. Tucker, A. Trexler, A. van Loon, Do we need a social media accelerator?, SocArXiv doi 10 (2023)

    2023

  31. [39]

    Reiter-Haas, E

    M. Reiter-Haas, E. Lex, The framing loop: Do users repeatedly read similar framed news online?, in: IUI Workshops, volume 3660, CEUR-WS.org, 2024

    2024

  32. [40]

    Bednar, Polarization, diversity, and democratic robustness, Proceedings of the National Academy of Sciences 118 (2021) e2113843118

    J. Bednar, Polarization, diversity, and democratic robustness, Proceedings of the National Academy of Sciences 118 (2021) e2113843118

    2021

  33. [41]

    Kotkov, S

    D. Kotkov, S. Wang, J. Veijalainen, A survey of serendipity in recommender systems, Knowledge- Based Systems 111 (2016) 180–192

    2016

Pith tools

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