Pith. sign in

REVIEW 4 major objections 4 minor 56 references

Fuzzy Norm-Explicit Product Quantization for Recommender Systems

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims that codebooks built from Interval Type-2 fuzzy clustering, fused by a Sugeno integral, improve the recall of norm-explicit product quantization for maximum inner product search on recommender benchmarks while keeping…

desk verdict A plausible fuzzy-clustering extension of NEQ with modest reported recall gains, but an undefined Sugeno fusion step and broken references keep it from being reproducible as submitted. read the letter →

arxiv 2412.06069 v1 pith:V3CTNQXM submitted 2024-12-08 cs.IR cs.AI

Mohammadreza Jamalifard , Javier Andreu-Perez , Hani Hagras , Luis Martínez López This is my paper · ORCID
classification cs.IRcs.AI
keywords productquantizationnorm-explicitmaximuminnersearchfuzzyclusteringtype-2setsSugenointegralrecommendersystemsrecall
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 tries to establish that product quantization for maximum inner product search, used in recommender systems to find the highest-scoring user-item matches, can be made more accurate by building its codebooks with soft fuzzy clustering instead of hard K-means. Its proposed method, Fuzzy-2 NEQ, replaces K-means in Norm-Explicit Product Quantization with Interval Type-2 Fuzzy Possibilistic C-Means and fuses the resulting fuzzy codebooks with a Sugeno integral, so each sub-vector can belong to more than one codeword. On the Netflix, Audio, and Cifar60k benchmarks, it reports recall of 94%, 69%, and 59%, respectively, outperforming NEQ, PQ, and RQ by up to +6%, +5%, and +8% at the larger item settings. The authors argue that the gain matters because recall improves without a meaningful increase in running time, keeping the method suitable for high-traffic services where query speed and low cost are priorities. If the central claim is right, soft codebook construction is a low-cost way to push the recall of quantization-based recommender retrieval beyond what hard clustering achieves.

What carries the argument

The load-bearing machinery is the fuzzy codebook plus fusion step. Interval Type-2 Fuzzy Possibilistic C-Means (IT2FPCM) is a soft clustering algorithm in which each data point has interval-valued memberships in several clusters, so the codebook codewords become type-2 fuzzy sets instead of hard centroids. The Sugeno integral, an ordinal fuzzy aggregation rule, combines those fuzzy codebooks into the single crisp codebook required by the rest of the norm-explicit pipeline. Around this, the method keeps NEQ's separation of each vector into a scalar norm and a unit direction, encoding the direction with the fuzzy-derived codebooks and the relative norm with separate scalar codebooks. The paper's claim is that this combination lowers quantization error at cluster boundaries and thereby raises recall per item retrieved.

What would settle it

Run Fuzzy-2 NEQ on Netflix, Audio, and Cifar60k with the Sugeno-fused fuzzy codebooks replaced by K-means centroids while keeping every other algorithm step identical; if recall at the larger item counts does not drop, the reported gains are not caused by the fuzzy codebook construction.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that fuzzifying the codebook construction step of Norm-Explicit Product Quantization yields better recall on maximum inner product search without sacrificing the method's efficiency. In Fuzzy-2 NEQ, the norm and the direction of each item vector are quantized separately, as in NEQ, but the direction codebooks are trained with a type-2 fuzzy possibilistic clustering algorithm rather than K-means. Because the clustering output is a fuzzy tensor rather than a crisp set of centroids, the method aggregates it with a Sugeno integral into a single crisp codebook before encoding. The paper reports that this procedure satisfies Lloyd's optimality conditions and, in experiments across Netflix, Audio, and Cifar60k, is statistically more accurate than NEQ, PQ, OPQ, RQ, Norm-RQ, AQ, APQ, and a simple locality-sensitive hashing baseline, while its running time stays close to that of plain PQ.

Load-bearing premise

The whole advantage rests on the premise that Sugeno-fused type-2 fuzzy clusters form a codebook that quantizes item vectors better than K-means centroids, and this premise is not independently established because the paper never specifies the fuzzy measure in the Sugeno integral.

Editorial extensions

If this is right

  • If the reported gains hold, recommender systems that already use product quantization can raise recall by swapping K-means for the fuzzy clustering and Sugeno fusion step, with essentially no change in query-time cost.
  • The method leaves the MIPS-to-nearest-neighbor transformation untouched, so it can be inserted into existing PQ-based retrieval pipelines without changing the distance computation.
  • At the largest tested settings, Fuzzy-2 NEQ's recall advantage appears without increasing the number of codebooks, implying the memory footprint of the compressed index stays comparable to PQ.
  • The recall improvements on Cifar60k (image embeddings) and Audio (voice embeddings) suggest the benefit is not specific to user-item rating data, if the paper's benchmark results are representative.

Reading between the lines

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

  • The paper does not state the fuzzy measure $g$ used in Eq. (12), so a natural next experiment would be to test whether the recall gain survives under different fuzzy measures or under a simpler weighted average; this is an editorial suggestion, not a result in the paper.
  • Because the fuzzy clustering step is independent of the norm-explicit decomposition, the same fuzzy codebook construction could plausibly be applied to other vector quantizers such as OPQ, RQ, or AQ; the paper only tests it inside NEQ.
  • The paper asserts that the Sugeno integral outperforms the Choquet integral for this fusion but reports no direct comparison; a head-to-head aggregation comparison would clarify whether the ordinal nature of Sugeno is essential.
  • If soft codebooks genuinely help because boundary vectors straddle clusters, then datasets with more cluster overlap should show larger gains; checking whether the gain magnitude correlates with cluster overlap would be a testable extension.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes Fuzzy-2 NEQ, a variant of Norm-Explicit Product Quantization in which the vector codebooks are trained with Interval Type-2 Fuzzy Possibilistic C-Means and the resulting fuzzy codebooks are aggregated with a Sugeno integral. The authors report recall improvements over PQ, OPQ, AQ, RQ, NormRQ, APQ, Simple Hash, and NEQ on Netflix, Audio, and Cifar60k, while claiming that the running time and complexity remain close to the most efficient existing PQ method. The manuscript includes the NEQ background, the proposed two-step codebook construction, experimental tables (Tables II-VI), and an appendix describing the genetic-algorithm tuning of the two fuzzy exponents.

Significance. If the proposed fuzzy codebook construction were fully specified and reproducible, it would offer a low-cost way to replace hard K-means codebooks in MIPS-oriented product quantization, which is a relevant direction for large-scale recommender systems. The paper deserves credit for comparing against seven baselines on three benchmark datasets and for reporting precision/F1 in addition to recall. However, the central aggregation mechanism is currently not defined precisely enough to be implemented, and the experimental evaluation does not provide a valid uncertainty measure for the claimed recall gains. As written, the contribution is therefore not yet verifiable.

major comments (4)
  1. [Section IV-B, Eq. (12), and Algorithm 2] Equation (12) does not define the Sugeno integral used in Algorithm 2, Step 3. The fuzzy measure g is never specified, h is only described as "cluster centers," and the domain and range of the integral are unclear. The standard discrete Sugeno integral requires a set function g on the index set of the fuzzy codebooks, and different choices of g will produce different aggregated crisp codewords and therefore different recall values. The statement that "the product has been used as a t-norm of Sugeno integral" is also inconsistent with the standard min-max formulation of the Sugeno integral. Consequently, the step that converts the fuzzy codebooks into the crisp codebook consumed by the rest of NEQ is not implementable as written, and the reported recall improvements cannot yet be reproduced or attributed to the proposed method.
  2. [Section V, Tables II and IV] The abstract's headline numbers are selected as the largest values in Table II (94.65 for Netflix at 16,384 items, 69.71 for Audio at 32,768 items, and 59.34 for Cifar60k at 32,768 items), so they do not describe a single experimental setting, and the baseline corresponding to each of the claimed '+6%, +5%, +8%' gains is not clearly indicated. Table IV is described in the text as reporting the standard deviation of recall, but the table actually reports only time standard deviations, and Table II contains no error bars for recall. Without an uncertainty measure, the claimed advantages over NEQ, PQ, and RQ are not statistically supported.
  3. [Section V and Appendix B] The fuzzy exponents ξ1=8.5 and ξ2=9.1 are tuned with a genetic algorithm on the same benchmark datasets that are later used for the reported recall, and the number of codebooks and clusters is also chosen per benchmark. Because these parameters directly determine the fuzzy partition and hence the aggregated codebook, this tuning protocol does not establish that the method generalizes. A held-out tuning split or a sensitivity analysis over the parameter grid shown in Figure 7 is needed before the improvements can be attributed to the fuzzy mechanism rather than to dataset-specific hyperparameter selection.
  4. [Section V, Tables IV and V] The abstract claims that the computing time and complexity of Fuzzy-2 NEQ "nearly equals" the most computationally efficient existing PQ method, but Table IV contains no NEQ row, and Table V reports CPU and memory usage for NEQ rather than running time. The method used as the reference point for the time claim is therefore not actually compared in the timing table. Please report the training and query time of NEQ alongside the other baselines, or temper the complexity claim.
minor comments (4)
  1. [Section V-A, Eq. (14)] The precision formula divides by "All Relevant Documents," which is the same denominator used for recall in Eq. (13); precision should divide by the number of retrieved documents.
  2. [Algorithm 1] The second loop is written as "while s = m′+1 to s = m," which is not a well-formed while condition; it should be a for loop, and the return statement should make clear that l·r is computed after both loops terminate.
  3. [Section II and References [15]-[17]] Several related-work citations do not match the methods they are meant to support: [15] is not the original product-quantization paper of Jegou et al., and [16] does not correspond to the OPQ work described in the text; the same issue affects the references for RQ and AQ. Please correct the bibliography so that the baseline descriptions can be verified.
  4. [Section V, Table IV] The table header is confusing: "Average Time Std Time Std" does not match the text's statement that the table shows the average standard deviation of recall. The header and caption should be aligned with the values actually tabulated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's central claim is an empirical comparison against external PQ baselines, and no load-bearing step reduces by construction to its own inputs.

full rationale

This is an empirical systems paper rather than a formal derivation, and the central claim is that replacing K-means codebook training in NEQ with Interval Type-2 Fuzzy Possibilistic C-means plus Sugeno-integral aggregation improves recall on Netflix, Audio, and Cifar60k relative to published baselines (PQ, OPQ, RQ, AQ, NEQ, NormRQ, APQ, Simple Hash). The comparison is externally grounded in the reported tables, so the main conclusion is not equivalent to its inputs. I inspected each load-bearing step for circularity. The fuzzy codebook construction is justified by citing Rubio et al. (Ref. [35]), an external source with no author overlap with the present paper, so this is not a self-citation chain. The Sugeno fusion step is under-specified: Eq. (12) writes an integral over fuzzy codebooks with respect to a fuzzy measure g, but g is never defined, and the statement that 'product has been used as a t-norm of Sugeno integral' is non-standard; this is a reproducibility gap, not a circular reduction, because the reported recall values are not forced by the formula. The genetic-algorithm tuning of the fuzzy exponents xi_1 and xi_2 and the selection of the largest item-count columns in Table II are in-sample selection and overfitting concerns, and the abstract quotes the largest observed differences; however, the paper reports retrospective evaluation rather than relabeling a fitted parameter as a prediction of a closely related quantity, so this does not satisfy the fitted-input-called-prediction pattern. The only self-citations ([50]-[55]) appear in future-work suggestions and are not load-bearing for the central result. No equation equates the output to the input, no fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. Therefore the appropriate finding is no significant circularity.

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

The central method rests on a cited fuzzy clustering algorithm and an underspecified aggregation operator. The only new math in the paper is a generic existence proof for optimal codebooks in Appendix A, which does not constrain the fuzzy construction. The two fuzzy exponents and codebook settings are tuned on the evaluation data, so the central claim depends on several free parameters.

free parameters (3)
  • xi1 and xi2 (fuzziness and possibility exponents for IT2FPCM) = 8.5 and 9.1
    Tuned by genetic algorithm on the benchmark datasets themselves. The values appear in Section III-B and Appendix B, and the reported recall depends on them.
  • Number of codebooks and clusters per experiment = Settings include 4-CB/4-Cl, 8-CB/32-Cl, 16-CB/32-Cl, 32-CB/100-Cl, 128-CB/100-Cl
    These are chosen per dataset and setting in Table VI, and the best-performing settings are used for the headline recall numbers.
  • Sugeno integral fuzzy measure g = not specified
    Eq. (12) uses a fuzzy measure g without defining it. The aggregation result depends on this unspecified measure, making the method underspecified.
assumptions (3)
  • domain assumption Interval Type-2 Fuzzy Possibilistic C-means produces a near-optimal codebook satisfying Lloyd's optimality conditions.
    Invoked in Section IV-A with citation [35], but no proof or verification is given for the NEQ setting in this paper.
  • ad hoc to paper The Sugeno integral with some undisplayed fuzzy measure is the correct way to fuse fuzzy codebooks into a crisp codebook.
    The paper asserts this in Section IV-B, but the measure g is never defined and no comparison justifies the choice over other aggregation operators.
  • domain assumption Norm-explicit quantization reduces norm error and improves MIPS performance as established in [13].
    The paper builds on NEQ as a given; this assumption is standard in the cited prior work and is not re-derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fuzzy Norm-Explicit Product Quantization for Recommender Systems." pith.science (2026). https://pith.science/paper/V3CTNQXM

@misc{pith2026241206069,
  author       = {Pith},
  title        = {Pith review of: Fuzzy Norm-Explicit Product Quantization for Recommender Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V3CTNQXM}},
  note         = {Machine review of arXiv:2412.06069}
}
read the original abstract

As the data resources grow, providing recommendations that best meet the demands has become a vital requirement in business and life to overcome the information overload problem. However, building a system suggesting relevant recommendations has always been a point of debate. One of the most cost-efficient techniques in terms of producing relevant recommendations at a low complexity is Product Quantization (PQ). PQ approaches have continued developing in recent years. This system's crucial challenge is improving product quantization performance in terms of recall measures without compromising its complexity. This makes the algorithm suitable for problems that require a greater number of potentially relevant items without disregarding others, at high-speed and low-cost to keep up with traffic. This is the case of online shops where the recommendations for the purpose are important, although customers can be susceptible to scoping other products. This research proposes a fuzzy approach to perform norm-based product quantization. Type-2 Fuzzy sets (T2FSs) define the codebook allowing sub-vectors (T2FSs) to be associated with more than one element of the codebook, and next, its norm calculus is resolved by means of integration. Our method finesses the recall measure up, making the algorithm suitable for problems that require querying at most possible potential relevant items without disregarding others. The proposed method outperforms all PQ approaches such as NEQ, PQ, and RQ up to +6%, +5%, and +8% by achieving a recall of 94%, 69%, 59% in Netflix, Audio, Cifar60k datasets, respectively. More and over, computing time and complexity nearly equals the most computationally efficient existing PQ method in the state-of-the-art.

Figures

Figures reproduced from arXiv: 2412.06069 by the authors.

Figure 1
Figure 1. The figure above represents the stages of the algorithm [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Time comparison of proposed the fuzzy method vs [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 5
Figure 5. Netflix dataset comparison plot [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: This figure presents a graphical comparison of the recall [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: This figure shows a grid of cost function values for [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 46 canonical work pages

  1. [15]

    Evaluating token-level and passage- level dense retrieval models for math information retrieval,

    W. Zhong, J.-H. Yang, and J. Lin, “Evaluating token-level and passage- level dense retrieval models for math information retrieval,” arXiv preprint arXiv:2203.11163, 2022

    arXiv 2022

  2. [16]

    Semantic models for the first-stage retrieval: A comprehensive review,

    J. Guo, Y . Cai, Y . Fan, F. Sun, R. Zhang, and X. Cheng, “Semantic models for the first-stage retrieval: A comprehensive review,” ACM Transactions on Information Systems (TOIS) , vol. 40, no. 4, pp. 1–42, 2022

    2022

  3. [17]

    A comprehensive survey on trustworthy recommender systems,

    W. Fan, X. Zhao, X. Chen, J. Su, J. Gao, L. Wang, Q. Liu, Y . Wang, H. Xu, L. Chen et al. , “A comprehensive survey on trustworthy recommender systems,” arXiv preprint arXiv:2209.10117 , 2022

    arXiv 2022

  4. [18]

    If beam search is the answer, what was the question?

    C. Meister, T. Vieira, and R. Cotterell, “If beam search is the answer, what was the question?” arXiv preprint arXiv:2010.02650 , 2020

    arXiv 2010

  5. [12]

    A survey on stream- based recommender systems,

    M. Al-Ghossein, T. Abdessalem, and A. Barre, “A survey on stream- based recommender systems,” ACM computing surveys (CSUR), vol. 54, no. 5, pp. 1–36, 2021

    work page 2021

  6. [1]

    Different but equal? a field experiment on the impact of recommendation systems on mobile and personal computer channels in retail,

    D. Lee, A. Gopal, and S.-H. Park, “Different but equal? a field experiment on the impact of recommendation systems on mobile and personal computer channels in retail,” Information Systems Research , vol. 31, no. 3, pp. 892–912, 2020

    work page 2020

  7. [2]

    A survey of recommendation systems: recommendation models, techniques, and application fields,

    H. Ko, S. Lee, Y . Park, and A. Choi, “A survey of recommendation systems: recommendation models, techniques, and application fields,” Electronics, vol. 11, no. 1, p. 141, 2022

    work page 2022

  8. [3]

    A review of modern fashion recommender systems,

    Y . Deldjoo, F. Nazary, A. Ramisa, J. Mcauley, G. Pellegrini, A. Bellogin, and T. Di Noia, “A review of modern fashion recommender systems,” arXiv preprint arXiv:2202.02757 , 2022. 11

    arXiv 2022

Show all 56 references
  1. [4]

    Recommender systems in the healthcare domain: state-of-the-art and research issues,

    T. N. T. Tran, A. Felfernig, C. Trattner, and A. Holzinger, “Recommender systems in the healthcare domain: state-of-the-art and research issues,” Journal of Intelligent Information Systems , vol. 57, no. 1, pp. 171–201, 2021

    2021

  2. [5]

    Learning tree-based deep model for recommender systems,

    H. Zhu, X. Li, P. Zhang, G. Li, J. He, H. Li, and K. Gai, “Learning tree-based deep model for recommender systems,” in Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , 2018, pp. 1079–1088

    2018

  3. [6]

    Efficient retrieval of matrix factorization-based top-k recommendations: A survey of recent approaches,

    D. D. Le and H. Lauw, “Efficient retrieval of matrix factorization-based top-k recommendations: A survey of recent approaches,” Journal of Artificial Intelligence Research , vol. 70, pp. 1441–1479, 2021

    2021

  4. [7]

    A survey of quantization methods for efficient neural network inference,

    A. Gholami, S. Kim, Z. Dong, Z. Yao, M. W. Mahoney, and K. Keutzer, “A survey of quantization methods for efficient neural network inference,” arXiv preprint arXiv:2103.13630 , 2021

    2021 arXiv

  5. [8]

    Recommender systems survey,

    J. Bobadilla, F. Ortega, A. Hernando, and A. Guti ´errez, “Recommender systems survey,” Knowledge-based systems, vol. 46, pp. 109–132, 2013

    2013

  6. [9]

    Fuzzy tools in recommender systems: A survey,

    R. Yera and L. Martinez, “Fuzzy tools in recommender systems: A survey,” International Journal of Computational Intelligence Systems , vol. 10, no. 1, p. 776, 2017

    2017

  7. [10]

    Recommender system application developments: a survey,

    J. Lu, D. Wu, M. Mao, W. Wang, and G. Zhang, “Recommender system application developments: a survey,” Decision support systems , vol. 74, pp. 12–32, 2015

    2015

  8. [11]

    Content- based group recommender systems: A general taxonomy and further improvements,

    Y . P´erez-Almaguer, R. Yera, A. A. Alzahrani, and L. Mart´ınez, “Content- based group recommender systems: A general taxonomy and further improvements,” Expert Systems with Applications , vol. 184, p. 115444, 2021

    2021

  9. [13]

    Norm- explicit quantization: Improving vector quantization for maximum inner product search,

    X. Dai, X. Yan, K. K. Ng, J. Liu, and J. Cheng, “Norm- explicit quantization: Improving vector quantization for maximum inner product search,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 01, 2020, pp. 51–58

    2020

  10. [14]

    Vector quantization: a review,

    Z.-b. Wu and J.-q. Yu, “Vector quantization: a review,” Frontiers of Information Technology & Electronic Engineering , vol. 20, no. 4, pp. 507–524, 2019

    2019

  11. [19]

    A shape-gain approach for vector quantization based on flat tori

    F. Boaventura de Miranda and C. Torezzan, “A shape-gain approach for vector quantization based on flat tori.” Advances in Mathematics of Communications, vol. 14, no. 3, 2020

    2020

  12. [20]

    Advances and open problems in federated learning,

    P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings et al. , “Advances and open problems in federated learning,” Foundations and Trends® in Machine Learning , vol. 14, no. 1–2, pp. 1–210, 2021

    2021

  13. [21]

    A survey on conversational recommender systems,

    D. Jannach, A. Manzoor, W. Cai, and L. Chen, “A survey on conversational recommender systems,” ACM Computing Surveys (CSUR), vol. 54, no. 5, pp. 1–36, 2021

    2021

  14. [22]

    Recommender systems: Techniques, applications, and challenges,

    F. Ricci, L. Rokach, and B. Shapira, “Recommender systems: Techniques, applications, and challenges,” Recommender Systems Handbook, pp. 1–35, 2022

    2022

  15. [23]

    Information retrieval models for contact recommendation in social networks,

    J. Sanz-Cruzado and P. Castells, “Information retrieval models for contact recommendation in social networks,” in Advances in Information Retrieval: 41st European Conference on IR Research, ECIR 2019, Cologne, Germany, April 14–18, 2019, Proceedings, Part I 41. Springer, 2019,...

    2019

  16. [24]

    The state-of-the-art in expert recommendation systems,

    N. Nikzad-Khasmakhi, M. Balafar, and M. R. Feizi-Derakhshi, “The state-of-the-art in expert recommendation systems,” Engineering Applications of Artificial Intelligence , vol. 82, pp. 126–147, 2019

    2019

  17. [25]

    Automated expertise retrieval: a taxonomy-based survey and open issues,

    R. Gonc ¸alves and C. F. Dorneles, “Automated expertise retrieval: a taxonomy-based survey and open issues,” ACM Computing Surveys (CSUR), vol. 52, no. 5, pp. 1–30, 2019

    2019

  18. [26]

    Virtual customer assistants in finance: From state of the art and practices to design guidelines,

    A. Iovine, F. Narducci, C. Musto, M. de Gemmis, and G. Semeraro, “Virtual customer assistants in finance: From state of the art and practices to design guidelines,” Computer Science Review , vol. 47, p. 100534, 2023

    2023

  19. [27]

    Predicting cancer drug response using a recommender system,

    C. Suphavilai, D. Bertrand, and N. Nagarajan, “Predicting cancer drug response using a recommender system,” Bioinformatics, vol. 34, no. 22, pp. 3907–3914, 2018

    2018

  20. [28]

    Applying optimal weight combination in hybrid recommender systems

    N. Haubner and T. Setzer, “Applying optimal weight combination in hybrid recommender systems.” in HICSS, 2020, pp. 1–10

    2020

  21. [29]

    Scalability and sparsity issues in recommender datasets: a survey,

    M. Singh, “Scalability and sparsity issues in recommender datasets: a survey,” Knowledge and Information Systems , vol. 62, pp. 1–43, 2020

    2020

  22. [30]

    A survey of autoencoder-based recommender systems,

    G. Zhang, Y . Liu, and X. Jin, “A survey of autoencoder-based recommender systems,” Frontiers of Computer Science, vol. 14, pp. 430– 450, 2020

    2020

  23. [31]

    A survey of active learning in collaborative filtering recommender systems,

    M. Elahi, F. Ricci, and N. Rubens, “A survey of active learning in collaborative filtering recommender systems,”Computer Science Review, vol. 20, pp. 29–50, 2016

    2016

  24. [32]

    Human-in-the-loop machine learning: a state of the art,

    E. Mosqueira-Rey, E. Hern ´andez-Pereira, D. Alonso-R ´ıos, J. Bobes- Bascar´an, and ´A. Fern ´andez-Leal, “Human-in-the-loop machine learning: a state of the art,” Artificial Intelligence Review , pp. 1–50, 2022

    2022

  25. [33]

    Solving separation-of- concerns problems in collaborative design of human-ai systems through leaky abstractions,

    H. Subramonyam, J. Im, C. Seifert, and E. Adar, “Solving separation-of- concerns problems in collaborative design of human-ai systems through leaky abstractions,” in Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems , 2022, pp. 1–21

    2022

  26. [34]

    Speeding up the xbox recommender system using a euclidean transformation for inner- product spaces,

    Y . Bachrach, Y . Finkelstein, R. Gilad-Bachrach, L. Katzir, N. Koenigstein, N. Nice, and U. Paquet, “Speeding up the xbox recommender system using a euclidean transformation for inner- product spaces,” in Proceedings of the 8th ACM Conference on Recommender systems, 2014, pp. 257–264

    2014

  27. [35]

    Interval type-2 fuzzy possibilistic c- means clustering algorithm,

    E. Rubio, O. Castillo, and P. Melin, “Interval type-2 fuzzy possibilistic c- means clustering algorithm,” in Recent Developments and New Direction in Soft-Computing Foundations and Applications . Springer, 2016, pp. 185–194

    2016

  28. [36]

    Interval type-2 fuzzy systems,

    J. M. Mendel, “Interval type-2 fuzzy systems,” in Uncertain Rule-Based Fuzzy Systems: Introduction and New Directions, 2nd Edition. Springer, 2017, pp. 449–527

    2017

  29. [37]

    Hyperparameter optimization: Comparing genetic algorithm against grid search and bayesian optimization,

    H. Alibrahim and S. A. Ludwig, “Hyperparameter optimization: Comparing genetic algorithm against grid search and bayesian optimization,” in 2021 IEEE Congress on Evolutionary Computation (CEC). IEEE, 2021, pp. 1551–1559

    2021

  30. [38]

    Quantization/clustering: when and why does k-means work?

    C. Levrard, “Quantization/clustering: when and why does k-means work?” Journal de la soci ´et´e franc ¸aise de statistique, vol. 159, no. 1, pp. 1–26, 2018

    2018

  31. [39]

    Nonlinearity of the fuzzy integral,

    E. P. Klement and D. Ralescu, “Nonlinearity of the fuzzy integral,” Fuzzy sets and systems , vol. 11, no. 1, pp. 309–315, 1983

    1983

  32. [40]

    Qualitative integrals and cointegrals: A survey,

    A. Rico, “Qualitative integrals and cointegrals: A survey,” in New Trends in Aggregation Theory 10 . Springer, 2019, pp. 3–14

    2019

  33. [41]

    Sugeno integral generalization applied to improve adaptive image binarization,

    F. Bardozzo, B. De La Osa, L. Horansk ´a, J. Fumanal-Idocin, M. delli Priscoli, L. Troiano, R. Tagliaferri, J. Fernandez, and H. Bustince, “Sugeno integral generalization applied to improve adaptive image binarization,” Information Fusion, vol. 68, pp. 37–45, 2021

    2021

  34. [42]

    Data fusion: theory, methods, and applications,

    M. Gagolewski, “Data fusion: theory, methods, and applications,” arXiv preprint arXiv:2208.01644, 2022

    2022 arXiv

  35. [43]

    Motor-imagery-based brain–computer interface using signal derivation and aggregation functions,

    J. Fumanal-Idocin, Y .-K. Wang, C.-T. Lin, J. Fern ´andez, J. A. Sanz, and H. Bustince, “Motor-imagery-based brain–computer interface using signal derivation and aggregation functions,” IEEE Transactions on Cybernetics, vol. 52, no. 8, pp. 7944–7955, 2021

    2021

  36. [44]

    C. L. Borgman, D. Moghdam, and P. K. Corbett, Effective online searching: A basic text . CRC Press, 2021

    2021

  37. [45]

    Challenges for recommender systems evaluation,

    F. Ricci, D. Massimo, and A. De Angeli, “Challenges for recommender systems evaluation,” in CHItaly 2021: 14th Biannual Conference of the Italian SIGCHI Chapter , 2021, pp. 1–5

    2021

  38. [46]

    Empirical study: visual analytics for comparing stacking to blending ensemble learning,

    A. Chatzimparmpas, R. M. Martins, K. Kucher, and A. Kerren, “Empirical study: visual analytics for comparing stacking to blending ensemble learning,” in 2021 23rd International Conference on Control Systems and Computer Science (CSCS) . IEEE, 2021, pp. 1–8

    2021

  39. [47]

    Precision vs. recall – an intuitive guide for every machine learning person,

    P. Huilgol, “Precision vs. recall – an intuitive guide for every machine learning person,” https://www.analyticsvidhya.com/blog/2020/ 09/precision-recall-machine-learning, Sep. 2020, accessed: 2022-8-17

    2020

  40. [48]

    A troubling analysis of reproducibility and progress in recommender systems research,

    M. Ferrari Dacrema, S. Boglio, P. Cremonesi, and D. Jannach, “A troubling analysis of reproducibility and progress in recommender systems research,” ACM Transactions on Information Systems (TOIS) , vol. 39, no. 2, pp. 1–49, 2021

    2021

  41. [49]

    K-means clustering and vector quantization (scipy.cluster.vq) — SciPy v0.8.dev reference guide (DRAFT),

    “K-means clustering and vector quantization (scipy.cluster.vq) — SciPy v0.8.dev reference guide (DRAFT),” http://library.isr.ist.utl.pt/ docs/scipy/cluster.vq.html, accessed: 2022-8-17

    2022

  42. [50]

    Towards understanding human functional brain development with explainable artificial intelligence: Challenges and perspectives,

    M. Kiani, J. Andreu-Perez, H. Hagras, S. Rigato, and M. L. Filippetti, “Towards understanding human functional brain development with explainable artificial intelligence: Challenges and perspectives,” IEEE Computational Intelligence Magazine , vol. 17, no. 1, pp. 16–33, 2022. 12

    2022

  43. [51]

    Effective brain connectivity for fnirs with fuzzy cognitive maps in neuroergonomics,

    M. Kiani, J. Andreu-Perez, H. Hagras, E. I. Papageorgiou, M. Prasad, and C.-T. Lin, “Effective brain connectivity for fnirs with fuzzy cognitive maps in neuroergonomics,” IEEE Transactions on Cognitive and Developmental Systems , vol. 14, no. 1, pp. 50–63, 2019

    2019

  44. [52]

    Real time recognition of human activities from wearable sensors by evolving classifiers,

    J. Andreu, R. D. Baruah, and P. Angelov, “Real time recognition of human activities from wearable sensors by evolving classifiers,” in 2011 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2011) . IEEE, 2011, pp. 2786–2793

    2011

  45. [53]

    A gentle introduction and survey on computing with words (cww) methodologies,

    P. K. Gupta and J. Andreu-Perez, “A gentle introduction and survey on computing with words (cww) methodologies,” Neurocomputing, vol. 500, pp. 921–937, 2022

    2022

  46. [54]

    Deep learning towards intelligent vehicle fault diagnosis,

    M. Al-Zeyadi, J. Andreu-Perez, H. Hagras, C. Royce, D. Smith, P. Rzonsowski, and A. Malik, “Deep learning towards intelligent vehicle fault diagnosis,” in 2020 International Joint Conference on Neural Networks (IJCNN). IEEE, 2020, pp. 1–7

    2020

  47. [55]

    Single-trial recognition of video gamer’s expertise from brain haemodynamic and facial emotion responses,

    A. R. Andreu-Perez, M. Kiani, J. Andreu-Perez, P. Reddy, J. Andreu- Abela, M. Pinto, and K. Izzetoglu, “Single-trial recognition of video gamer’s expertise from brain haemodynamic and facial emotion responses,” Brain sciences, vol. 11, no. 1, p. 106, 2021. Mohammadreza Jamalif...

    2022

  48. [1970]

    and Ph.D

    He received the M.Sc. and Ph.D. degrees in Computer Sciences, both from the University of Granada, Spain, in 1993 and 1999, respectively. Currently, he is a Full Professor in Computer Science Department at the University of Ja´en. He is the main researcher in the Research Grou...

    1993

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.