Pith. sign in

REVIEW 3 major objections 4 minor 56 references

StARS: Socially Appropriate Robot Actions via a Recommender System-Driven Approach

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A collaborative-filtering layer added to any scene encoder lets a robot predict how appropriate each action is for a specific user, and the gains over averaged human ratings are consistent across two datasets and eleven backbones.

desk verdict Useful empirical demonstration that recommender-style personalization improves social appropriateness prediction, but the mechanism claim is undercut by a missing bias-only ablation and cold-start evaluation. read the letter →

arxiv 2607.21802 v1 pith:67FBNFMB submitted 2026-07-23 cs.RO cs.IR

Erencem Ozbey , Fethiye Irmak Dogan , Jin Huang , Hatice Gunes This is my paper · ORCID
classification cs.ROcs.IR
keywords sociallyappropriaterobotactionspersonalizationcollaborativefilteringrecommendersystemsmatrixfactorizationhuman-robotinteractionsocialnavigationinter-subjectvariability
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 argues that social appropriateness in human-robot interaction is not universal: the same robot action in the same scene can be judged appropriate by one person and inappropriate by another. It reformulates the problem as a recommender-system task, treating annotators as users, scenes as items, and appropriateness scores for candidate actions as ratings. The proposed StARS framework augments any differentiable scene encoder with a collaborative-filtering layer based on matrix factorization, learning user, scene, and action embeddings jointly so each user receives personalized scores. Across two datasets and eleven backbone models, StARS consistently reduces prediction error and improves agreement with human raters, with statistically reliable gains even under sparse preference feedback.

What carries the argument

Multi-task matrix factorization with residual fusion. User and scene embeddings are shared across actions while each action has its own embedding; the hybrid scene embedding is the scene encoder's content embedding plus a collaborative scene embedding, and the personalized score is the dot product of an action embedding with the Hadamard product of the user embedding and hybrid scene embedding, plus a residual content term and user, scene, and global offsets. This combines collaborative signal (who rates what) with content signal (scene features), and the residual content shortcut stabilizes prediction when collaborative factors are weak.

What would settle it

Take either dataset, split by annotator identity rather than by row so no annotator appears in both training and test, and retrain StARS; if the gains over aggregated baselines disappear or reverse, the personalisation effect does not generalize to unseen users. A second check: give the model a cold-start user with only a few ratings and see whether it still beats an aggregated baseline that ignores identity.

Watch

Extended reading notes

Core claim

StARS treats social appropriateness as a user-item rating problem and shows that collaborative filtering can be layered on top of any differentiable scene encoder without redesigning it. The prediction for a user, scene, and action is a dot product between an action embedding and a user-personalized scene representation, where the scene representation is the sum of a collaborative scene embedding and a content embedding from the scene encoder, plus a residual content term and offset terms. This structure lets the model share statistical strength across users and scenes while still outputting user-specific scores. Evaluated on MannersDB+ and SocNav1, StARS outperforms the scene-only version o

Load-bearing premise

The reported gains assume the target user is already known when training starts: ratings are split randomly, so the same individuals appear in training and test, and every personalised score requires a user embedding.

Editorial extensions

If this is right

  • Socially appropriate robot action selection can be personalized without retraining or replacing the underlying perception or decision model; the collaborative-filtering layer is an add-on.
  • The improvements are systematic, not architecture-specific: all eleven backbone-dataset pairs show the same direction of gain.
  • Personalization works in sparse regimes: meaningful gains appear with a fraction of the data, which matters because collecting per-user ratings is costly.
  • The structure of inter-subject disagreement is learnable: rating variation correlates with user-scene interactions, not just noise.
  • Recommender-system methods are transferable to HRI preference modelling, opening a path to rank alternative robot actions by user-specific appropriateness.

Reading between the lines

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

  • If the method is evaluated with a user-wise split (no overlap of annotators between training and test), the gains may shrink or vanish, because the model must then infer a user embedding from scratch; the paper only reports random-split results.
  • The same architecture could be applied to other personalization-centric HRI tasks, such as preferred assistance levels in collaborative assembly or preferred communication style, treating the user as the 'user' and interaction contexts as 'items'.
  • A testable extension: use cold-start users (no prior ratings) and initialize their embeddings from scene-only predictions; if aggregated baselines then become competitive, the value of collaborative filtering lies in warm-start personalization.
  • The residual content shortcut suggests the model is robust when collaborative factors are weak; one could intentionally ablate the residual term to measure its contribution.
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 / 4 minor

Summary. The paper reformulates socially appropriate robot action prediction as a personalised rating prediction problem. StARS augments a differentiable scene encoder with a multi-task matrix-factorization collaborative-filtering (CF) module: Eq. (5) combines shared user/scene embeddings, action embeddings, user/scene/global offsets, and residual content fusion. Training is two-stage (scene-encoder pre-training, then end-to-end joint optimisation). The method is evaluated on MannersDB+ with five scene-only backbones and on SocNav1 with six GNN backbones, using RMSE/MSE, Pearson r and CCC, with Bouckaert-Frank-corrected paired t-tests and Holm adjustment. Tables II and III report consistent StARS improvements over the scene-only baselines in all 11 backbone-dataset pairs with small adjusted p-values, and Figure 3 reports learning curves for two StARS variants under varying training fractions.

Significance. If the empirical claims hold, StARS is a useful model-agnostic personalisation layer for socially appropriate robot action selection, and the paper's public code, two datasets, eleven backbones, and corrected significance testing are genuine strengths. However, the manuscript does not yet establish that the CF interaction term, rather than simple per-user offsets, is responsible for the reported gains, and the evaluation protocol assumes that target users are already known at training time. These gaps are load-bearing for the paper's central claim that collaborative filtering is the mechanism and that the method supports practical personalised deployment.

major comments (3)
  1. [§III-B, Eq. (5); Tables II and III] The central claim attributes the improvement to a collaborative-filtering module, but Eq. (5) bundles several user-specific components: the offset b_u, the scene offset b_i, the content residual h_i^T r_a, and the interaction term (p_u ⊙ (q_i + h_i))^T r_a. The scene-only baseline contains no user identifier, so any user-specific signal, including a simple per-user mean offset, would reduce error. With MannersDB+ having only 20.8 annotations per annotator on average, a per-user intercept could plausibly capture much of the variance. The paper never ablates the interaction term. Please add a bias-only variant (e.g., yhat = h_i^T r_a + b_u + b_i + b, without p_u ⊙ qbar_i) and report the same metrics and tests. If such a variant recovers most of the RMSE/Pearson/CCC improvement, the conclusion should be weakened from 'collaborative filtering helps' to 'adding user IDs helps'.
  2. [§IV-D, Eq. (5); Discussion] The evaluation splits the full set of annotated user–scene rows randomly into train and test, so the same annotators appear in both splits. Because Eq. (5) requires a user embedding p_u for every prediction, this protocol does not measure performance for new users with no prior ratings (cold start). The Discussion claims real-world deployment where the model is 'initialised from offline annotations' and adapts user embeddings, but no user-wise split or cold-start experiment is reported. Please add an evaluation where some users are held out entirely, or otherwise state clearly that the reported gains apply only to known users with existing ratings. This is necessary to support the deployment-facing claims in Section VI.
  3. [§V-C, Fig. 3; RQ3] The RQ3 claim that the RS formulation is 'particularly beneficial' or 'particularly effective' in the low-data regime is not supported by the presented learning curves. Figure 3 shows only StARS performance as a function of training fraction, with no comparison to the corresponding scene-only baseline or to a bias-only variant at the same fractions. The curves could simply reflect the backbone's own data-scaling behaviour. Please include the scene-only baseline (and ideally the bias-only ablation) in Figure 3, and report significance or confidence intervals for the gap; otherwise the sparsity-robustness conclusion is unsubstantiated.
minor comments (4)
  1. [§IV-E] Typo: 'across-validation–corrected paired t-test' should read 'cross-validation–corrected paired t-test'.
  2. [§III-B, Eq. (5)] The notation is a bit dense: h_i appears both as part of the hybrid scene embedding qbar_i = q_i + h_i and as a separate residual term. A one-sentence explanation of why the second h_i is needed, and how it differs from the first, would improve readability.
  3. [Algorithm 1, §IV-D] The algorithm says z-score statistics are computed 'from training split D', while the text says normalization is applied per training split. Please clarify whether mu_a and sigma_a are recomputed in each outer fold and whether the inner validation split is included in these statistics.
  4. [Fig. 3] The learning curves show only five-fold means without error bars or pointwise variability. Adding confidence intervals or per-fold points would make the claimed diminishing-returns trend easier to assess.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation loop: StARS is an empirical regression model evaluated on held-out annotations. The only mild flag is a self-citation ([21]) used for conceptual motivation, not for any load-bearing step.

full rationale

StARS is an empirical predictive model trained with a standard regression loss (Eq. 1) and evaluated on held-out user–scene–action ratings (Section IV-D). The personalized score in Eq. (5) is not derived from the target or from a fitted quantity renamed as a prediction; the test rows are distinct from training rows, and user embeddings are latent parameters inferred from training annotations, which is standard collaborative-filtering evaluation. The reported improvements over scene-only backbones are therefore an empirical comparison, not a tautology. The missing bias-only ablation noted by the skeptic concerns attribution of the gain to per-user offsets versus interaction terms; that is an experimental gap, not circularity. The only self-citation is [21] (Huang et al., HRI 2026), cited in Section II-C as a 'conceptual framework' motivating recommender systems for social robots. It is not used to justify any equation, result, or uniqueness claim, and it does not carry the paper's empirical conclusions. Because the evaluation is external (MannersDB+, SocNav1) and the code is released, the central claim is self-contained. Score 2 reflects the minor non-load-bearing self-citation; no circular step is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The framework rests on standard collaborative-filtering assumptions rather than physical axioms. No new physical entities are introduced; the user, scene, and action latent factors are learned model parameters, not independent theoretical postulates. The most consequential unstated premise is transductive evaluation: user embeddings are assumed available at test time because the same users appear in train and test.

free parameters (3)
  • Latent embedding dimension d = not reported in paper
    Dimension of user, scene, and action embeddings in the MF module; controls model capacity and is chosen by hand rather than by a reported search.
  • MF parameters p_u, q_i, r_a, b_u, b_i, b = learned via AdamW; values not reported
    These fitted latent factors and biases are the mechanism by which StARS personalizes scores; they are the core learned parameters of the framework.
  • Per-action z-score normalization mu_a, sigma_a = computed per training split; values not reported
    Target scores are normalized per action using training-split statistics and inverted at evaluation; this changes the target scale and affects all reported errors.
assumptions (4)
  • domain assumption Appropriateness scores can be approximated by low-rank interaction of user and scene latent factors plus a content residual (Eqs. 2 and 5).
    This is the core collaborative-filtering assumption; no derivation is given, and the paper relies on empirical improvement to justify it.
  • domain assumption Row-wise random splitting of user-scene annotations is a valid test of the deployment claim, i.e., every test user has a learned training embedding.
    Section IV-D splits user-scene rows without a user-wise partition, and Eq. (5) requires p_u; cold-start users are therefore outside the evaluated regime.
  • domain assumption Scene encoder pre-training with a temporary linear head transfers to the joint personalized model.
    Algorithm 1 Step 1 uses this two-stage transfer; it is a standard but unproven assumption in this setting.
  • domain assumption MSE on normalized per-action scores is an adequate surrogate for the quality of final robot action selection.
    Eq. (1) instantiates the loss as MSE; the paper does not evaluate a downstream action-selection metric or user-facing outcome.

how reviews work

0 comments
Cite this review

Pith. "Pith review of StARS: Socially Appropriate Robot Actions via a Recommender System-Driven Approach." pith.science (2026). https://pith.science/paper/67FBNFMB

@misc{pith2026260721802,
  author       = {Pith},
  title        = {Pith review of: StARS: Socially Appropriate Robot Actions via a Recommender System-Driven Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/67FBNFMB}},
  note         = {Machine review of arXiv:2607.21802}
}
read the original abstract

Social appropriateness in human-robot interaction (HRI) is not universal: different people can judge the same robot action differently in the same situation. To capture this inter-subject variability, we reformulate socially appropriate action generation as a preference modelling problem inspired by recommender systems, treating annotators as users, contexts/scenes as items, and appropriateness scores over a set of candidate robot actions as targets. We propose StARS, a novel model-agnostic framework that integrates collaborative filtering with learnable scene representations to generate user-specific appropriateness scores over candidate robot actions. StARS is model-agnostic: it can be integrated with various scene encoders and backbones, enabling personalisation without redesigning the underlying model. We evaluate StARS on two socially aware robotics datasets, MannersDB+ and SocNav1, and analyse robustness under sparse preference feedback. Across datasets and backbones, StARS consistently improves performance and agreement with annotators, supporting personalised action selection aligned with user norms. Our code is publicly available at https://github.com/Cambridge-AFAR/StARS.git.

Figures

Figures reproduced from arXiv: 2607.21802 by the authors.

Figure 1
Figure 1. Social appropriateness depends on personalised pref [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall pipeline of the StARS framework. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Scaling behavior with increasing training data. Macro-averaged RMSE/MSE (five-fold mean) and agreement metrics (Pearson’s r and CCC) are plotted against the fraction of the training dataset used. expect the RS formulation to be particularly beneficial in low-data regimes where each user provides few ratings. Fig￾ure 3a reports a learning-curve on MannersDB+ for the FT￾Transformer-StARS model, where performance is ag… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 6 linked inside Pith

  1. [1]

    C. C. Aggarwal,Recommender systems. Springer, 2016

    2016

  2. [2]

    Reinforcement learning approaches in social robotics,

    N. Akalinet al., “Reinforcement learning approaches in social robotics,”Sensors, vol. 21, no. 4, p. 1292, 2021

    2021

  3. [3]

    Evaluating the replicability of significance tests for comparing learning algorithms,

    R. R. Bouckaertet al., “Evaluating the replicability of significance tests for comparing learning algorithms,” inProc. PAKDD. Springer, 2004, pp. 3–12

    2004

  4. [4]

    Should i follow the crowd? a probabilistic analysis of the effectiveness of popularity in recommender systems,

    R. Ca ˜namareset al., “Should i follow the crowd? a probabilistic analysis of the effectiveness of popularity in recommender systems,” inProc. ACM SIGIR, 2018, pp. 415–424

    2018

  5. [5]

    Socially aware motion planning with deep reinforcement learning,

    Y . F. Chen,et al., “Socially aware motion planning with deep reinforcement learning,” inProc. IROS, Sept. 2017, pp. 1343–1350

    2017

  6. [6]

    Can robots earn our trust the same way humans do? a systematic exploration of competence, warmth, and anthropomorphism as determinants of trust development in hri,

    L. Christoforakos,et al., “Can robots earn our trust the same way humans do? a systematic exploration of competence, warmth, and anthropomorphism as determinants of trust development in hri,”Front. Robot. AI, vol. 8, p. 640444, 2021

    2021

  7. [7]

    Feature aggregation with latent generative replay for federated continual learning of socially appropriate robot behaviours,

    N. Churamani,et al., “Feature aggregation with latent generative replay for federated continual learning of socially appropriate robot behaviours,”arXiv preprint arXiv:2405.15773, 2024

    arXiv 2024

  8. [8]

    Long-term personalization of an in-home socially assistive robot for children with autism spectrum disorders,

    C. Clabaugh,et al., “Long-term personalization of an in-home socially assistive robot for children with autism spectrum disorders,”Front. Robot. AI, vol. 6, p. 110, 2019

    2019

Show all 56 references
  1. [9]

    Think, act, and ask: Open-world interactive personal- ized robot navigation,

    Y . Dai,et al., “Think, act, and ask: Open-world interactive personal- ized robot navigation,” inProc. ICRA. IEEE, 2024, pp. 3296–3303

    2024

  2. [10]

    Learning personalized human-aware robot navigation using virtual reality demonstrations from a user study,

    J. De Heuvel,et al., “Learning personalized human-aware robot navigation using virtual reality demonstrations from a user study,” in Proc. RO-MAN. IEEE, 2022, pp. 898–905

    2022

  3. [11]

    Learning depth vision-based personalized robot navigation from dynamic demonstrations in virtual reality,

    J. De Heuvel,et al., “Learning depth vision-based personalized robot navigation from dynamic demonstrations in virtual reality,” inProc. IROS. IEEE, 2023, pp. 6757–6764

    2023

  4. [12]

    Towards proactive social robots: Distilling visual knowledge from large vision-language models,

    G. De Simone,et al., “Towards proactive social robots: Distilling visual knowledge from large vision-language models,” inProc. ICCV, 2025, pp. 2492–2502

    2025

  5. [13]

    Grace: Generating socially appropriate robot actions leveraging llms and human explanations,

    F. I. Do ˘gan,et al., “Grace: Generating socially appropriate robot actions leveraging llms and human explanations,” inProc. ICRA. IEEE, 2025, pp. 4330–4336

    2025

  6. [14]

    Are we really making much progress? a worrying analysis of recent neural recommendation approaches,

    M. Ferrari Dacrema,et al., “Are we really making much progress? a worrying analysis of recent neural recommendation approaches,” in Proc. RecSys, 2019, pp. 101–109

    2019

  7. [15]

    Learning Socially Appropriate Robot Approaching Behavior Toward Groups using Deep Reinforcement Learning,

    Y . Gao,et al., “Learning Socially Appropriate Robot Approaching Behavior Toward Groups using Deep Reinforcement Learning,” in Proc. RO-MAN, Oct. 2019, pp. 1–8

    2019

  8. [16]

    Revisiting deep learning models for tabular data,

    Y . Gorishniy,et al., “Revisiting deep learning models for tabular data,” Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 34, pp. 18 932–18 943, 2021

    2021

  9. [17]

    Federated continual learning for socially aware robotics,

    L. Guerdanet al., “Federated continual learning for socially aware robotics,” inProc. RO-MAN. IEEE, 2023, pp. 1522–1529

    2023

  10. [18]

    Inductive representation learning on large graphs,

    W. Hamilton,et al., “Inductive representation learning on large graphs,”Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 30, 2017

    2017

  11. [19]

    Neural collaborative filtering,

    X. He,et al., “Neural collaborative filtering,” inProc. WWW, 2017, pp. 173–182

    2017

  12. [20]

    Personalization and localization in human-robot interaction: A review of technical methods,

    M. Hellou,et al., “Personalization and localization in human-robot interaction: A review of technical methods,”Robotics, vol. 10, no. 4, p. 120, 2021

    2021

  13. [21]

    Reimagining social robots as recommender systems: Foundations, framework, and applications,

    J. Huang,et al., “Reimagining social robots as recommender systems: Foundations, framework, and applications,” inProc. HRI, 2026

    2026

  14. [22]

    Personalization in long-term human-robot interaction,

    B. Irfan,et al., “Personalization in long-term human-robot interaction,” inProc. HRI. IEEE, 2019, pp. 685–686

    2019

  15. [23]

    Recommender systems: Past, present, future,

    D. Jannach,et al., “Recommender systems: Past, present, future,”AI Mag., vol. 42, pp. 3–6, 2021

    2021

  16. [24]

    Socially CompliAnt Navigation Dataset (SCAND): A Large-Scale Dataset of Demonstrations for Social Navigation,

    H. Karnan,et al., “Socially CompliAnt Navigation Dataset (SCAND): A Large-Scale Dataset of Demonstrations for Social Navigation,”IEEE Robot. Autom. Lett., vol. 7, no. 4, pp. 11 807–11 814, Oct. 2022

    2022

  17. [25]

    Understanding large-language model (llm)-powered human-robot interaction,

    C. Y . Kim,et al., “Understanding large-language model (llm)-powered human-robot interaction,” inProc. HRI, 2024, pp. 371–380

    2024

  18. [26]

    Semi-supervised classification with graph convolu- tional networks,

    T. N. Kipfet al., “Semi-supervised classification with graph convolu- tional networks,”arXiv preprint arXiv:1609.02907, 2016

    2016 arXiv

  19. [27]

    Matrix factorization techniques for recommender systems,

    Y . Koren,et al., “Matrix factorization techniques for recommender systems,”Computer, vol. 42, no. 8, pp. 30–37, 2009

    2009

  20. [28]

    Advances in collaborative filtering,

    Y . Koren,et al., “Advances in collaborative filtering,”Recommender Syst. Handb., pp. 91–142, 2021

    2021

  21. [29]

    Analyzing the effects of human-aware motion planning on close-proximity human–robot collaboration,

    P. A. Lasotaet al., “Analyzing the effects of human-aware motion planning on close-proximity human–robot collaboration,”Hum. Fac- tors, vol. 57, no. 1, pp. 21–33, 2015

    2015

  22. [30]

    Examining the impact of robot norm viola- tions on participants’ trust, discomfort, behaviour and physiological responses—a mixed method approach,

    S. Lawrence,et al., “Examining the impact of robot norm viola- tions on participants’ trust, discomfort, behaviour and physiological responses—a mixed method approach,”ACM Trans. Hum.-Robot Interact., vol. 14, no. 3, pp. 1–32, 2025

    2025

  23. [31]

    Personalization in hri: A longitudinal field experi- ment,

    M. K. Lee,et al., “Personalization in hri: A longitudinal field experi- ment,” inProc. HRI, 2012, pp. 319–326

    2012

  24. [32]

    Gated graph sequence neural networks,

    Y . Li,et al., “Gated graph sequence neural networks,”arXiv preprint arXiv:1511.05493, 2015

    2015 arXiv

  25. [33]

    A review on interactive reinforcement learning from human social feedback,

    J. Lin,et al., “A review on interactive reinforcement learning from human social feedback,”IEEE Access, vol. 8, pp. 120 757–120 765, 2020

    2020

  26. [34]

    Decoupled weight decay regularization,

    I. Loshchilovet al., “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  27. [35]

    Recommender systems,

    L. L ¨u,et al., “Recommender systems,”Phys. Rep., vol. 519, no. 1, pp. 1–49, 2012

    2012

  28. [36]

    Explanations and trust: What happens to trust when a robot partner does something unexpected?

    J. B. Lyons,et al., “Explanations and trust: What happens to trust when a robot partner does something unexpected?”Comput. Hum. Behav., vol. 138, p. 107473, 2023

    2023

  29. [37]

    Graph neural networks for human-aware social navigation,

    L. J. Manso,et al., “Graph neural networks for human-aware social navigation,” inProc. WAF. Springer, 2020, pp. 167–179

    2020

  30. [38]

    Socnav1: A dataset to benchmark and learn social navigation conventions,

    L. J. Manso,et al., “Socnav1: A dataset to benchmark and learn social navigation conventions,”Data, vol. 5, no. 1, p. 7, 2020

    2020

  31. [39]

    Optimization of criterion for objective evaluation of hri performance that approximates subjective evaluation: a case study in robot competition,

    Y . Mizuchiet al., “Optimization of criterion for objective evaluation of hri performance that approximates subjective evaluation: a case study in robot competition,”Adv. Robot., vol. 34, no. 3-4, pp. 142–156, 2020

    2020

  32. [40]

    Designing social robots with llms for engag- ing human interaction,

    M. Pinto-Bernal,et al., “Designing social robots with llms for engag- ing human interaction,”Appl. Sci., vol. 15, no. 11, p. 6377, 2025

    2025

  33. [41]

    Robot gains social intelligence through mul- timodal deep reinforcement learning,

    A. H. Qureshi,et al., “Robot gains social intelligence through mul- timodal deep reinforcement learning,” inProc. Humanoids. IEEE, 2016, pp. 745–751

    2016

  34. [42]

    Bpr: Bayesian personalized ranking from implicit feedback,

    S. Rendle,et al., “Bpr: Bayesian personalized ranking from implicit feedback,” inProc. UAI, 2009, pp. 452–461

    2009

  35. [43]

    Adapting a robot’s linguistic style based on socially-aware reinforcement learning,

    H. Ritschel,et al., “Adapting a robot’s linguistic style based on socially-aware reinforcement learning,” inProc. RO-MAN. IEEE, 2017, pp. 378–384

    2017

  36. [44]

    Preface to the special issue on personalization and adaptation in human–robot interactive communication: S. rossi et al

    S. Rossi,et al., “Preface to the special issue on personalization and adaptation in human–robot interactive communication: S. rossi et al.” User Model. User-Adapt. Interact., vol. 33, no. 2, pp. 189–194, 2023

    2023

  37. [45]

    Collaborative filtering recommender systems,

    J. B. Schafer,et al., “Collaborative filtering recommender systems,” inThe Adaptive Web. Springer, 2007, pp. 291–324

    2007

  38. [46]

    Modeling relational data with graph convo- lutional networks,

    M. Schlichtkrull,et al., “Modeling relational data with graph convo- lutional networks,” inProc. ESWC. Springer, 2018, pp. 593–607

    2018

  39. [47]

    A survey of collaborative filtering techniques,

    X. Suet al., “A survey of collaborative filtering techniques,”Adv. Artif. Intell., vol. 2009, no. 1, p. 421425, 2009

    2009

  40. [48]

    Beyond algorithms: An hci perspective on recommender systems,

    K. Swearingenet al., “Beyond algorithms: An hci perspective on recommender systems,” inProc. ACM SIGIR Workshop, vol. 13, no. 5-6, 2001, pp. 1–11

    2001

  41. [49]

    Mind your manners! a dataset and a contin- ual learning approach for assessing social appropriateness of robot actions,

    J. Tjomsland,et al., “Mind your manners! a dataset and a contin- ual learning approach for assessing social appropriateness of robot actions,”Front. Robot. AI, vol. 9, p. 669420, 2022

    2022

  42. [50]

    SEAN: Social Environment for Autonomous Naviga- tion,

    N. Tsoi,et al., “SEAN: Social Environment for Autonomous Naviga- tion,” inProc. HAI, ser. HAI ’20. New York, NY , USA: Association for Computing Machinery, Nov. 2020, pp. 281–283

    2020

  43. [51]

    Graph attention networks,

    P. Veli ˇckovi´c,et al., “Graph attention networks,”arXiv preprint arXiv:1710.10903, 2017

    2017 arXiv

  44. [52]

    Extracting and composing robust features with denoising autoencoders,

    P. Vincent,et al., “Extracting and composing robust features with denoising autoencoders,” inProc. ICML, 2008, pp. 1096–1103

    2008

  45. [53]

    Robotic etiquette: Results from user studies involving a fetch and carry task,

    M. L. Walters,et al., “Robotic etiquette: Results from user studies involving a fetch and carry task,” inProc. HRI, ser. HRI ’07. New York, NY , USA: Association for Computing Machinery, Mar. 2007, pp. 317–324

    2007

  46. [54]

    Personalization in human-robot interaction through preference-based action representation learning,

    R. Wang,et al., “Personalization in human-robot interaction through preference-based action representation learning,” inProc. ICRA. IEEE, 2025, pp. 7377–7384

    2025

  47. [55]

    Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems,

    R. Wang,et al., “Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems,” inProc. WWW, 2021, pp. 1785–1797

    2021

  48. [56]

    How powerful are graph neural networks?

    K. Xu,et al., “How powerful are graph neural networks?”arXiv preprint arXiv:1810.00826, 2018

    2018 arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.