Pith. sign in

REVIEW 4 major objections 4 minor 43 references

Prink: $k_s$-Anonymization for Streaming Data in Apache Flink

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Prink extends k_s-anonymization to categorical and hierarchical streaming data inside Apache Flink.

desk verdict Prink is a genuine, code-backed step toward practical categorical stream anonymization in Flink, but the distributed guarantee is unproven and the evaluation is thinner than the claims. read the letter →

arxiv 2505.13153 v1 pith:OWKAJ3AJ submitted 2025-05-19 cs.DC cs.CRcs.SE

Philip Groneberg , Saskia Nuñez von Voigt , Thomas Janke , Louis Loechel , Karl Wolf , Elias Grünewald , Frank Pallas This is my paper · ORCID
classification cs.DCcs.CRcs.SE
keywords k_s-anonymitystreamingdataanonymizationApacheFlinknon-numericalgeneralizationhierarchyinformationlossl-diversityCASTLE
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 claims that k_s-anonymity, previously limited mostly to numerical streaming data, can be applied to categorical and hierarchically structured attributes in a streaming pipeline, and that this can be done inside Apache Flink with acceptable performance and information loss. The authors present Prink, a Flink-native implementation that builds on the CASTLE clustering algorithm and adds dynamic generalization hierarchies, semantics-aware loss metrics, support for multiple sensitive attributes, and distribution across Flink nodes. If the claims hold, Prink would give real-world stream processing systems a practical way to anonymize non-numerical personal data while still emitting discrete, usable tuples rather than aggregates. The result matters because many production streams, such as energy, traffic, or health data, contain categorical and hierarchical fields that existing stream anonymizers either ignore or coarsen too aggressively.

What carries the argument

The load-bearing mechanism is the CASTLE clustering loop extended with dynamic domain generalization hierarchies: each cluster maintains its own DGH built and updated from the stream, and generalization cost is measured by the Generalized Loss Metric (GLM), Normalized Certainty Penalty (NCP), or Per Record Loss (PRL), which weights frequent values more heavily. The enlargement value, i.e., the added information loss a new tuple would cause if assigned to a candidate cluster, decides cluster placement, and the delay constraint delta forces release of the oldest cluster after delta tuples. The Flink-specific machinery is a KeyedBroadcastProcessFunction that broadcasts CastleRule objects for runtime rule changes while keying tuples to their data subjects.

What would settle it

Run Prink with Flink parallelism greater than one on a stream where each node receives fewer than k distinct data subjects (for example, two nodes, k=10, with only five distinct subjects routed to one node), and inspect the released clusters: if any released cluster contains fewer than k distinct subject identifiers, the distributed k_s-anonymity claim is false.

Watch

Extended reading notes

Core claim

The central claim is that semantics-aware k_s-anonymization of non-numerical streaming data is not only possible but practically deployable: Prink generalizes each incoming tuple via per-cluster dynamic domain generalization hierarchies (DGHs), assigns tuples to clusters by an enlargement value computed from semantic loss metrics (GLM, NCP, or the frequency-aware PRL), enforces l-diversity over one or more sensitive attributes, and releases generalized clusters under a delay constraint delta. The paper further claims that this can run natively in Apache Flink through a single KeyedBroadcastProcessFunction, that distribution is possible while preserving k_s and l guarantees by keying tuples to their data subjects, and that experiments on the ASHRAE energy data set show the expected trade-off between k, delta, latency, and information loss, with higher delta lowering information loss at the cost of latency.

Load-bearing premise

The guarantee that every released cluster contains at least k distinct individuals assumes that, under parallel execution, every Flink node receives enough distinct data subjects to form a compliant cluster; if a node holds fewer than k distinct subjects, it cannot release any cluster without breaking the guarantee, and the paper offers no proof or configuration rule covering that case.

Editorial extensions

If this is right

  • Categorical and hierarchical attributes like country, workplace, or building use can be stream-anonymized while keeping k_s-anonymity and l-diversity guarantees.
  • Because Prink outputs discrete generalized tuples rather than aggregates, downstream analytics can run on anonymized data as if it were ordinary tuple data, enabling flexible reuse.
  • Setting delta as a tuning parameter gives operators a concrete dial between lower latency (small delta) and lower information loss (large delta).
  • Distribution across Flink nodes is possible in principle, with the caveat that each node must receive enough distinct data subjects to form compliant clusters.
  • The implementation pattern, a single ProcessFunction with broadcast rules, makes it easy to insert anonymization into an existing Flink pipeline at any point.

Reading between the lines

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

  • The k_s guarantee under parallel execution likely depends on the assumption that every Flink parallel subtask receives at least k distinct data subjects; if a node receives fewer, it may be forced to delay or drop tuples, and the paper offers no configuration guidance or proof for that case.
  • The per-cluster dynamic DGHs could cause the same individual to be generalized differently in different time windows, which the paper itself notes as a potential inconsistency risk; this suggests a testable extension where DGHs are synchronized or versioned across windows.
  • The PRL metric, which downweights frequent values, could be especially valuable in heavily skewed streams, and a natural extension would measure how much utility PRL preserves relative to GLM and NCP on skewed data.
  • The same architecture could plausibly extend to other enterprise streaming frameworks that support keyed broadcast state, since the anonymization logic itself is not Flink-specific.
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

4 major / 4 minor

Summary. The paper presents Prink, an Apache Flink-based implementation of ks-anonymization for streaming data, extending the CASTLE algorithm with support for categorical and hierarchical generalization, dynamic generalization hierarchies, and multiple sensitive attributes. The authors claim four contributions: semantics-aware ks-anonymization of non-numerical streaming data, an information-loss-optimized generalization scheme, a concept for distributing CASTLE's clustering across parallel nodes while preserving k- and l-guarantees, and an experimental evaluation on the ASHRAE energy data set measuring latency and information loss.

Significance. If the distributed guarantee claim is sound, Prink would be a practically relevant contribution: it is an open-source, Flink-native anonymizer that handles categorical and hierarchical data, with a detailed and reproducible bench-marking setup. The paper's strengths include a concrete implementation, a realistic data set, and an explicit discussion of parameter trade-offs. However, the central correctness claim for distributed execution is not proven, and the experimental evidence for 'acceptable' overheads and information loss lacks baselines and variance reporting. The contribution is therefore conditional on addressing these gaps.

major comments (4)
  1. [Section 3.1 and Section 4.1] The distributed ks-anonymity guarantee is not established. The paper states in Section 3.1 that keying tuples to data subjects is 'critical for preserving ks-anonymity and l-diversity guarantees' and supports parallel execution, but Flink's keyed partitioning only ensures all records for one subject go to the same subtask; it does not control how many distinct subjects land on a subtask. If parallelism p > N/k, the pigeonhole principle forces at least one subtask to receive fewer than k distinct subjects, so that subtask can never release a compliant cluster. Even for p <= N/k, hash-based key assignment can be imbalanced. Section 4.1 acknowledges only a scalability limitation, not this correctness/availability failure. The authors must either provide a proof that a subtask with fewer than k subjects can still release compliant clusters (e.g., via cross-partition coordination) or state and enforce a configuration constraint such as p <= N/k, and discuss the resulting availability implications.
  2. [Section 4.3 and Section 4.5] The experimental evaluation does not support the claim of 'acceptable performance overheads and information loss' because no baseline is reported. All latency figures are absolute end-to-end times (Figure 4) and internal processing times (Figure 5); without comparison to a non-anonymized pipeline or to the original CASTLE implementation, the reader cannot judge what overhead Prink introduces. Similarly, the information-loss results in Figure 3 have no reference point, such as CASTLE's behavior on the same data or a trivial generalization strategy. The authors should add at least one baseline comparison to substantiate the 'acceptable' claim.
  3. [Section 4.3 and Section 4.4] The evaluation reports averages from three runs per configuration but provides no error bars, standard deviations, or significance tests. Figures 3, 4, and 5 show only point estimates. Given that the paper describes 180 benchmark runs, variance information is essential to assess whether the observed trends (e.g., latency decreasing with increasing k) are reliable. The authors should report dispersion measures or at least include error bars for the key results.
  4. [Section 3.2 and Section 4.4] The information-loss evaluation is self-referential: the clustering algorithm optimizes cluster assignment using an enlargement value based on the Generalized Loss Metric (GLM), and the evaluation then measures information loss using the same GLM (Section 4.4). The reported 'information loss' is therefore partly a reflection of the objective function rather than an independent measure of utility. Reporting results with a different metric (e.g., NCP or PRL) or a downstream task (e.g., prediction accuracy on the anonymized data) would provide a more convincing utility assessment.
minor comments (4)
  1. [Abstract] The abstract contains a grammatical error: 'in a information loss-optimized manner' should be 'in an information loss-optimized manner'.
  2. [Figure 2 and Figure 3] The parameter ℓ is inconsistent between the experiment overview and the results. Figure 2 lists ℓ = {0, 2, 4, 8}, Figure 3 says 'with l = 1 held constant', and Figure 4 uses ℓ1, ℓ2, ℓ4, ℓ8. The authors should clarify which values of ℓ were used and why the figures differ.
  3. [Section 4.2] The sentence 'the building_id is used as the unique identifier providing ks-anonymity' is confusing because building_id is also listed in Table 1 as an attribute with a range and unique values. It should be clarified that building_id is the data-subject identifier (not a quasi-identifier), while the quasi-identifiers are the other attributes listed.
  4. [Equation (1)] In the definition of GLM, the notation M_p is not explicitly defined before use; it should state that M_p is the number of leaf nodes covered by the current generalization node u.

Circularity Check

1 steps flagged · score 2.0 of 10

Only mild self-referentiality: the information-loss metric is both the clustering objective and the evaluation measure; the privacy and latency claims remain externally grounded.

  1. other [Section 3.2 (Cluster Logic) and Section 4.4 (Evaluation Metrics), Eq. (1)]
    "assigning a data tuple to a cluster involves evaluating the information loss for each potential cluster and selecting the one with the lowest average information loss. ... Information Loss reduction is one of the core requirements of Prink. For the metric itself, the evaluation will use the information loss metrics explained in Section 3.2. For our evaluation, we use the Generalized Loss Metric for the calculation of the overall information loss."

    GLM, defined in Eq. (1), is simultaneously the objective minimized during clustering and the yardstick reported in the evaluation. Hence the information-loss results are not independent evidence for the 'information loss-optimized' claim; they restate the algorithm's own optimization target. This is a mild self-referentiality rather than a load-bearing circularity: it does not affect the k_s/l-diversity guarantee, the Flink integration, or the latency measurements.

full rationale

The central derivation is not circular. Prink builds on the externally published CASTLE algorithm for the core clustering and anonymity guarantees, and the paper's own contribution is an engineering adaptation to categorical and hierarchical data plus a Flink implementation. The k_s-anonymity and l-diversity checks are enforced directly by the cluster release logic, not derived from fitted values or from self-citations. The performance evaluation uses independently measured end-to-end latency and internal processing times. The only self-referential element is the evaluation of information loss: the paper uses GLM both as the metric that guides cluster assignment and as the metric reported in the results. This makes the 'low information loss' claim partly a restatement of the optimizer's objective, but it does not undermine the privacy guarantee or the system's practical functionality. The distributed-preservation concern is a correctness risk about Flink keying, not a circularity: the paper asserts that keying is 'critical for preserving k_s-anonymity and l-diversity' without proving that every parallel subtask receives enough distinct subjects, but this is an unproven condition rather than a reduction of the claim to its inputs.

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

No fitted free parameters or invented entities are introduced. The central claim rests on stream framework assumptions about keying and on the choice of loss metrics as utility proxies; those are listed as axioms.

assumptions (3)
  • domain assumption Flink's keyed partitioning routes all tuples of one data subject to the same node and distinct subjects can still be combined into clusters of size at least k on each node.
    Section 3.1 keys tuples to data subjects and claims this preserves k_s-anonymity; Section 4.1 acknowledges the scalability limit when few subjects exist. The parallel guarantee is not proven.
  • domain assumption The dynamic DGHs built from attribute values in the stream correctly represent the semantic relationships needed for meaningful generalization.
    Section 3.1 adds missing nodes to DGHs at runtime; if hierarchies are incomplete, loss metrics and generalizations are semantically wrong.
  • domain assumption Average GLM per cluster is an adequate proxy for data utility in the evaluation.
    Section 4.4 uses GLM as the sole utility metric; no downstream analytics are run to validate utility.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Prink: $k_s$-Anonymization for Streaming Data in Apache Flink." pith.science (2026). https://pith.science/paper/OWKAJ3AJ

@misc{pith2026250513153,
  author       = {Pith},
  title        = {Pith review of: Prink: $k_s$-Anonymization for Streaming Data in Apache Flink},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OWKAJ3AJ}},
  note         = {Machine review of arXiv:2505.13153}
}
read the original abstract

In this paper, we present Prink, a novel and practically applicable concept and fully implemented prototype for ks-anonymizing data streams in real-world application architectures. Building upon the pre-existing, yet rudimentary CASTLE scheme, Prink for the first time introduces semantics-aware ks-anonymization of non-numerical (such as categorical or hierarchically generalizable) streaming data in a information loss-optimized manner. In addition, it provides native integration into Apache Flink, one of the prevailing frameworks for enterprise-grade stream data processing in numerous application domains. Our contributions excel the previously established state of the art for the privacy guarantee-providing anonymization of streaming data in that they 1) allow to include non-numerical data in the anonymization process, 2) provide discrete datapoints instead of aggregates, thereby facilitating flexible data use, 3) are applicable in real-world system contexts with minimal integration efforts, and 4) are experimentally proven to raise acceptable performance overheads and information loss in realistic settings. With these characteristics, Prink provides an anonymization approach which is practically feasible for a broad variety of real-world, enterprise-grade stream processing applications and environments.

Figures

Figures reproduced from arXiv: 2505.13153 by the authors.

Figure 1
Figure 1. Prink Architecture in the context of Apache flink data streaming infrastructure. Prink’s architecture consists of the following core components, depicted in [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Benchmarking experiment architecture diagram 4.3 Experimental Setup We now proceed to outline the design considerations, configurations, and au￾tomation processes for the benchmarking experiment. An abstract overview is provided in [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Evaluation of accuracy through the information loss of attributes. tion, each configuration was tested three times, totaling 180 distinct benchmark￾ing runs. All code and deployment scripts for this evaluation are available in a dedicated GitHub repository7 . 4.4 Evaluation Metrics Information Loss Information loss reduction is one of the core requirements of Prink. For the metric itself, the evaluation will use the… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Evaluation of end-to-end latency, i.e., total time a data tuple spends within Prink As expected, a higher δ leads to increased latency, with a δ of 1250 resulting in the lowest latency and a δ of 20, 000 producing the highest. This occurs because δ directly affects the…
Figure 5
Figure 5. Figure 5: Evaluation of internal processing times, focusing on key functions: bestSelection, delayConstraint, and waitTime (the interval between these operations). 5 Discussion & Future Work In this section, we discuss the results in terms of the interaction between perfor￾mance…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 30 canonical work pages

  1. [1]

    In: SIGKDD ’18: Proceedings of the 24th ACM International Conference on Knowledge Discovery and Data Mining

    Abowd, J.M.: The us census bureau adopts differential privacy. In: SIGKDD ’18: Proceedings of the 24th ACM International Conference on Knowledge Discovery and Data Mining. pp. 2867–2867 (2018)

    work page 2018

  2. [2]

    Journal of Big Data5, 1–26 (2018)

    Al-Zobbi, M., Shahrestani, S., Ruan, C.: Experimenting sensitivity-based anonymization framework in apache spark. Journal of Big Data5, 1–26 (2018)

    work page 2018

  3. [3]

    In: CCS ’13: Proceedings of the 2013 ACM SIGSAC conference on Computer & communications security

    Andrés, M.E., Bordenabe, N.E., Chatzikokolakis, K., Palamidessi, C.: Geo- indistinguishability: Differential privacy for location-based systems. In: CCS ’13: Proceedings of the 2013 ACM SIGSAC conference on Computer & communications security. pp. 901–914 (2013)

    work page 2013

  4. [4]

    International Journal of Agricultural and Biosystems Engineering 5(9), 512–515 (2011)

    Apuan, D.A.: Landscape data transformation: Categorical descriptions to numeri- cal descriptors. International Journal of Agricultural and Biosystems Engineering 5(9), 512–515 (2011)

    work page 2011

  5. [5]

    Informa- tion Sciences 546, 1–24 (2021)

    Ashkouti, F., Sheikhahmadi, A., et al.: Di-mondrian: Distributed improved mon- drian for satisfaction of the l-diversity privacy model using apache spark. Informa- tion Sciences 546, 1–24 (2021)

    work page 2021

  6. [6]

    Nature com- munications 10(1), 4817 (2019)

    Bassolas, A., Barbosa-Filho, H., Dickinson, B., Dotiwalla, X., Eastham, P., Gal- lotti, R., Ghoshal, G., Gipson, B., Hazarie, S.A., Kautz, H., et al.: Hierarchical organization of urban mobility and its connection with city livability. Nature com- munications 10(1), 4817 (2019)

    work page 2019

  7. [7]

    Electronics10(5), 589 (2021)

    Bazai, S.U., Jang-Jaccard, J., Alavizadeh, H.: Scalable, high-performance, and gen- eralized subtree data anonymization approach for apache spark. Electronics10(5), 589 (2021)

    work page 2021

  8. [8]

    In: DPM ’23: Proceedings of 18th Inter- national Workshop on Data Privacy Management

    Brunn, C., Nuñez von Voigt, S., Tschorsch, F.: Analyzing continuous k s- anonymization for smart meter data. In: DPM ’23: Proceedings of 18th Inter- national Workshop on Data Privacy Management. Lecture Notes in Computer Science, vol. 14398, pp. 272–282. Springer (2023). https://doi.org/10.1007/978-3- 031-54204-6\_16

    doi:10.1007/978-3- 2023

Show all 43 references
  1. [9]

    IEEE Transactions on Dependable and Secure Computing8(3), 337– 352 (2010)

    Cao, J., Carminati, B., Ferrari, E., Tan, K.L.: Castle: Continuously anonymizing data streams. IEEE Transactions on Dependable and Secure Computing8(3), 337– 352 (2010)

    2010

  2. [10]

    In: SIGMOD ’18: Proceedings of the 2018 International Conference on Management of Data

    Cormode, G., Jha, S., Kulkarni, T., Li, N., Srivastava, D., Wang, T.: Privacy at scale: Local differential privacy in practice. In: SIGMOD ’18: Proceedings of the 2018 International Conference on Management of Data. pp. 1655–1658 (2018)

    2018

  3. [11]

    Advances in Neural Information Processing Systems30 (2017)

    Ding, B., Kulkarni, J., Yekhanin, S.: Collecting telemetry data privately. Advances in Neural Information Processing Systems30 (2017)

    2017

  4. [12]

    Computers & Mathematics with Applications38(11-12), 13–32 (1999) Prink: ks-Anonymization for Streaming Data in Apache Flink 21

    Domingo-Ferrer, J., Mateo-Sanz, J.M.: Resampling for statistical confidentiality in contingency tables. Computers & Mathematics with Applications38(11-12), 13–32 (1999) Prink: ks-Anonymization for Streaming Data in Apache Flink 21

    1999

  5. [13]

    In: ICALP’06: Proceedings of the 33rd Interna- tional Colloquium on Automata, Languages and Programming

    Dwork, C.: Differential privacy. In: ICALP’06: Proceedings of the 33rd Interna- tional Colloquium on Automata, Languages and Programming. pp. 1–12. Springer (2006). https://doi.org/10.1007/11787006_1

    2006 doi

  6. [14]

    Foundations and Trends® in Theoretical Computer Science 9(3–4), 211–407 (2014)

    Dwork, C., Roth, A., et al.: The algorithmic foundations of differential privacy. Foundations and Trends® in Theoretical Computer Science 9(3–4), 211–407 (2014)

    2014

  7. [15]

    Computer Science-Research and Development32, 173–182 (2017)

    Eibl, G., Engel, D.: Differential privacy for real smart metering data. Computer Science-Research and Development32, 173–182 (2017)

    2017

  8. [16]

    In: CCS ’14: Proceedings of the 21st ACM SIGSAC Conference on Computer and Communications Security

    Erlingsson, Ú., Pihur, V., Korolova, A.: Rappor: Randomized aggregatable privacy- preserving ordinal response. In: CCS ’14: Proceedings of the 21st ACM SIGSAC Conference on Computer and Communications Security. pp. 1054–1067. ACM (2014). https://doi.org/10.1145/2660267.2660348

    2014

  9. [17]

    In: CSITSS ’19: Proceedings of the 4th International Conference on Compu- tational Systems and Information Technology for Sustainable Solution

    Gadad, V., Sowmyarani, C.: A novel utility metric to measure information loss for generalization and suppression techniques in privacy preserving data publish- ing. In: CSITSS ’19: Proceedings of the 4th International Conference on Compu- tational Systems and Information Techn...

    2019

  10. [18]

    International Journal of Information Security and Privacy 2(3), 28–44 (2008)

    Gal,T.S.,Chen,Z.,Gangopadhyay,A.:Aprivacyprotectionmodelforpatientdata with multiple sensitive attributes. International Journal of Information Security and Privacy 2(3), 28–44 (2008). https://doi.org/10.4018/JISP.2008070103

    2008 doi

  11. [19]

    In: Big Data ’18: Proceedings of the 2018 IEEE International Conference on Big Data

    Gruschka,N.,Mavroeidis,V.,Vishi,K.,Jensen,M.:Privacyissuesanddataprotec- tion in big data: a case study analysis under gdpr. In: Big Data ’18: Proceedings of the 2018 IEEE International Conference on Big Data. pp. 5027–5033. IEEE (2018)

    2018

  12. [20]

    Nature communications 13(1), 29 (2022)

    Houssiau, F., Rocher, L., de Montjoye, Y.A.: On the difficulty of achieving differ- ential privacy in practice: user-level guarantees in aggregate location data. Nature communications 13(1), 29 (2022)

    2022

  13. [21]

    IEEE Transactions on Knowledge and Data Engineering 35(1), 108–127 (2021)

    Jiang, H., Pei, J., Yu, D., Yu, J., Gong, B., Cheng, X.: Applications of differential privacy in social network analysis: A survey. IEEE Transactions on Knowledge and Data Engineering 35(1), 108–127 (2021)

    2021

  14. [22]

    Kasiviswanathan, S.P., Lee, H.K., Nissim, K., Raskhodnikova, S., Smith, A.: What can we learn privately? SIAM Journal on Computing40(3), 793–826 (2011)

    2011

  15. [23]

    In: TCC ’13: Proceedings of the 10th Theory of Cryptography Conference

    Kasiviswanathan,S.P.,Nissim,K.,Raskhodnikova,S.,Smith,A.:Analyzinggraphs with node differential privacy. In: TCC ’13: Proceedings of the 10th Theory of Cryptography Conference. pp. 457–476. Springer (2013)

    2013

  16. [24]

    Computers & Security111, 102464 (2021)

    Kim, J.W., Edemacu, K., Kim, J.S., Chung, Y.D., Jang, B.: A survey of differ- ential privacy-based techniques and their applicability to location-based services. Computers & Security111, 102464 (2021)

    2021

  17. [25]

    In: ICDE ’07: Proceedings of the 23rd International Con- ference on Data Engineering

    Li, N., Li, T., Venkatasubramanian, S.: t-closeness: Privacy beyond k-anonymity and l-diversity. In: ICDE ’07: Proceedings of the 23rd International Con- ference on Data Engineering. pp. 106–115. IEEE Computer Society (2007). https://doi.org/10.1109/ICDE.2007.367856

    2007

  18. [26]

    In: ICDE ’06: Proceedings of the 22nd International Conference on Data Engineering

    Machanavajjhala, A., Gehrke, J., Kifer, D., Venkitasubramaniam, M.: l- Diversity: Privacy beyond k-Anonymity. In: ICDE ’06: Proceedings of the 22nd International Conference on Data Engineering. pp. 24–24 (2006). https://doi.org/10.1109/ICDE.2006.1

    2006 doi

  19. [27]

    IEEE access9, 8512–8545 (2020)

    Majeed, A., Lee, S.: Anonymization techniques for privacy preserving data pub- lishing: A comprehensive survey. IEEE access9, 8512–8545 (2020)

    2020

  20. [28]

    In: IC3K ’20: Proceedings of the 12th International Joint Conference on Knowl- edge Discovery, Knowledge Engineering and Knowledge Management

    Marques, J.F., Bernardino, J.: Analysis of data anonymization techniques. In: IC3K ’20: Proceedings of the 12th International Joint Conference on Knowl- edge Discovery, Knowledge Engineering and Knowledge Management. pp. 235–241. SCITEPRESS (2020). https://doi.org/10.5220/0010...

    2020 doi

  21. [29]

    In: IPMU ’12: Proceedings of the 14th International Con- ference on Information Processing and Management of Uncertainty in Knowledge- Based Systems

    Martínez, S., Sánchez, D., Valls, A.: Towards k-anonymous non-numerical data via semantic resampling. In: IPMU ’12: Proceedings of the 14th International Con- ference on Information Processing and Management of Uncertainty in Knowledge- Based Systems. Communications in Compute...

    2012 doi

  22. [30]

    Journal of Biomedical Informatics 46(2), 294–303 (2013)

    Martínez, S., Sánchez, D., Valls, A.: A semantic framework to pro- tect the privacy of electronic health records with non-numerical at- tributes. Journal of Biomedical Informatics 46(2), 294–303 (2013). https://doi.org/https://doi.org/10.1016/j.jbi.2012.11.005, https://www. sc...

    2013 doi

  23. [31]

    In: TPSIE ’19: Proceedings of the 1st Workshop on Trust and Privacy Aspects of Smart Information Environments

    Nuñez von Voigt, S., Tschorsch, F.: Rrtxfm: Probabilistic counting for differentially private statistics. In: TPSIE ’19: Proceedings of the 1st Workshop on Trust and Privacy Aspects of Smart Information Environments. vol. 573, pp. 86–98. Springer (2019). https://doi.org/10.100...

    2019 doi

  24. [32]

    Information Sciences467, 238–255 (2018)

    Otgonbayar, A., Pervez, Z., Dahal, K., Eager, S.: K-varp: K-anonymity for varied data streams via partitioning. Information Sciences467, 238–255 (2018)

    2018

  25. [33]

    IEEE Security & Privacy22(2), 86–92 (2024)

    Pallas, F., Koerner, K., Barberá, I., Hoepman, J.H., Jensen, M., Narla, N.R., Samarin, N., Ulbricht, M.R., Wagner, I., Wuyts, K., Zimmermann, C.: Privacy en- gineering from principles to practice: A roadmap. IEEE Security & Privacy22(2), 86–92 (2024). https://doi.org/10.1109/M...

    2024

  26. [34]

    In: M4IoT@Middleware 2021: Proceedings of the 8th International Workshop on Mid- dleware and Applications for the Internet of Things

    Pallas, F., Legler, J., Amslgruber, N., Grünewald, E.: Redcastle: practically applicable ks-anonymity for iot streaming data at the edge in node-red. In: M4IoT@Middleware 2021: Proceedings of the 8th International Workshop on Mid- dleware and Applications for the Internet of T...

    2021

  27. [35]

    In: ITAIC ’11: Proceedings of the 6th IEEE Joint International Information Technology and Artificial Intelligence Conference

    Qing-jiang, K., Xiao-hao, W., Jun, Z.: The ( p, α, k) anonymity model for privacy protection of personal information in the social networks. In: ITAIC ’11: Proceedings of the 6th IEEE Joint International Information Technology and Artificial Intelligence Conference. vol. 2, pp...

    2011

  28. [36]

    Robinson, A., Brown, F., Hall, N., Jackson, A., Kemp, G., Leeke, M.: Castleguard: Anonymised data streams with guaranteed differential privacy. In: 2020 IEEE Intl Conf on Dependable, Autonomic and Secure Com- puting, Intl Conf on Pervasive Intelligence and Computing, Intl Conf...

    2020

  29. [37]

    Samarati, P., Sweeney, L.: Protecting privacy when disclosing information: k- anonymity and its enforcement through generalization and suppression (1998)

    1998

  30. [38]

    IOP Conference Series: Materials Sci- ence and Engineering 225, 012279 (08 2017)

    Simi, M., Nayaki, K., Elayidom, M.: An extensive study on data anonymiza- tion algorithms based on k-anonymity. IOP Conference Series: Materials Sci- ence and Engineering 225, 012279 (08 2017). https://doi.org/10.1088/1757- 899X/225/1/012279

    2017 doi

  31. [39]

    In: Big Data ’17: Proceedings of the 2017 IEEE international conference on big data

    Sopaoglu, U., Abul, O.: A top-down k-anonymization implementation for apache spark. In: Big Data ’17: Proceedings of the 2017 IEEE international conference on big data. pp. 4513–4521. IEEE (2017)

    2017

  32. [40]

    In: ICIMIA ’20: Proceedings of the 2nd International Conference on Innovative Mechanisms for Industry Applications

    Suneetha, V., Suresh, S., Jhananie, V.: A novel framework using apache spark for privacy preservation of healthcare big data. In: ICIMIA ’20: Proceedings of the 2nd International Conference on Innovative Mechanisms for Industry Applications. pp. 743–749. IEEE (2020) Prink: ks-...

    2020

  33. [41]

    International journal of uncertainty, fuzziness and knowledge-based systems10(05), 557–570 (2002)

    Sweeney, L.: k-anonymity: A model for protecting privacy. International journal of uncertainty, fuzziness and knowledge-based systems10(05), 557–570 (2002)

    2002

  34. [42]

    ACM Com- puting Surveys (Csur)51(3), 1–38 (2018)

    Wagner, I., Eckhoff, D.: Technical privacy metrics: a systematic survey. ACM Com- puting Surveys (Csur)51(3), 1–38 (2018)

    2018

  35. [43]

    In: ICFFC ’10: Proceedings of the 2nd International Conference on Future Computer and Communication

    Zhang, J., Yang, J., Zhang, J., Yuan, Y.: Kids: K-anonymization data stream base on sliding window. In: ICFFC ’10: Proceedings of the 2nd International Conference on Future Computer and Communication. vol. 2, pp. V2–311. IEEE (2010)

    2010

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.