Decoding Complexity: Intelligent Pattern Exploration with CHPDA (Context Aware Hybrid Pattern Detection Algorithm)

Karanpreet Singh; Lokesh Koli; Shubham Kalra

REVIEW 5 major objections 4 minor 22 references

Decoding Complexity: Intelligent Pattern Exploration with CHPDA (Context Aware Hybrid Pattern Detection Algorithm)

T0 review · 5 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a hybrid pipeline combining Google RE2 regex matching, Aho-Corasick exact search, AI named entity recognition, and validation functions detects PII and PHI with 94–95% accuracy, a 91.6% F1 score, false positives…

desk verdict Central benchmark claims contradict the paper's own tables; the architecture is plausible but the evidence is not. read the letter →

arxiv 2502.07815 v1 pith:G4JH4CIQ submitted 2025-02-09 cs.CR cs.AI

Lokesh Koli , Shubham Kalra , Karanpreet Singh This is my paper · ORCID

classification cs.CRcs.AI

keywords PIIdetectionPHIGoogleRE2Aho-CorasickalgorithmnamedentityrecognitionhybridAIandregexdatalosspreventionconfidencescoring

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 sets out to show that sensitive data detection is best done by a hybrid: regular expressions for well-defined formats, an exact-match automaton for keywords, and AI named entity recognition for context, all coordinated by a confidence-scoring step. The authors claim their Context-Aware Hybrid Pattern Detection Algorithm (CHPDA) reaches about 94–95% accuracy on PII and PHI, an F1 score of 91.6%, false positives below 3%, and processing speeds near 100 MB/s on files up to 1 TB. A sympathetic reader would care because real-time data-loss prevention and compliance with GDPR, HIPAA, and CCPA depend on fast, low-false-positive detection in unstructured enterprise data. The paper also argues that Google RE2 is the best regex engine for this purpose and that Aho-Corasick is the best exact match algorithm, based on its own benchmarks.

What carries the argument

The load-bearing mechanism is the CHPDA pipeline: Google RE2's SET mechanism compiles many regex patterns into one deterministic finite automaton that scans text in a single linear pass; Aho-Corasick builds a trie with failure links to match a dictionary of context keywords simultaneously; a proximity scoring formula assigns a confidence $C_{\text{total}} = \text{Proximity\_Score}(d) + \text{Validation\_Score}$ based on the distance $d$ between a keyword and a detected pattern, with $\text{Proximity\_Score}(d) = \max(0, \alpha(D_{\max} - d))$; and a user-set threshold $T$ keeps only matches with $C \geq T$. Validation functions such as Luhn's algorithm for credit card numbers provide the validation score, which is what lowers false positives.

What would settle it

Re-run CHPDA on a publicly available labeled corpus of documents containing PII and PHI with the same pattern sets and measure F1 and false-positive rate; if the F1 is substantially below 91.6% or false positives exceed 3%, the central claim is not reproducible.

Watch

Extended reading notes

Core claim

The central claim is that combining rule-based and AI-based detection in one pipeline outperforms either approach alone: regex alone gives precision 92.5% but recall 75.3%, AI alone gives recall 89.2% but precision 84.7%, while the AI + Regex hybrid reaches precision 94.8% and recall 88.7%, yielding an F1 of 91.6% across dataset sizes from 100 MB to 5 GB. The paper further claims this hybrid maintains a stable confidence score and a false-positive rate below 3% at processing speeds around 100 MB/s on ordinary hardware.

Load-bearing premise

The evaluation numbers assume a controlled, representative test corpus with a fixed ground-truth set of PII/PHI labels, a published list of regex patterns, and repeated runs; the paper does not provide these details, so the reported 94–95% accuracy and 91.6% F1 could be artifacts of an ad hoc test.

Editorial extensions

If this is right

If the benchmark numbers hold, a single commodity server can scan a 1 TB corpus for PII/PHI in about three hours, making real-time DLP feasible for mid-size organizations.
The threshold $T$ gives operators a tunable precision-recall dial: raising it cuts false positives, lowering it catches more edge-case patterns.
The reported linear scaling of regex processing time with dataset size and pattern count means capacity planning can be a simple multiplicative estimate.
Because RE2 is preferred over Hyperscan for hardware compatibility, the pipeline can run on ARM-based lightweight agents and shared Windows environments, not just dedicated servers.

Reading between the lines

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

The proximity-scoring mechanism is portable: the same $C_{\text{total}}$ formula could be applied to any context-sensitive detection task, such as finding credentials in logs or sensitive clauses in contracts, not just PII/PHI.
Extending the NER stage with multilingual models is a natural next step: the paper identifies missing multilingual support as a limitation, and modern multilingual NER would likely close that gap without changing the pipeline.
The reported linear scaling could be stress-tested by adding exponentially larger pattern sets; the paper only tests up to 172 patterns, so the claim of linearity is an extrapolation beyond the measured range.

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.

Desk Editor's Note

read the letter

Short version: this is an engineering report with a plausible pipeline and a description of a hybrid detection approach, but the central numeric claims don't survive contact with the paper's own tables. The abstract's 100 MB/s and the memory figures are inconsistent with the timings in Table 3 and the memory in Table 7. The F1 claim in Table 4 also doesn't match the false-positive and missed-match counts in Table 5. These are not minor typos; they undermine the main results.

What's good: The explanation of why RE2 with its SET mechanism is a reasonable choice for multi-pattern regex matching is accurate and clear. The Aho-Corasick primer is correct, and the decision to combine regex with NER and validation functions is a sensible engineering direction. The proximity-confidence scoring idea, while simple, is a concrete way to weight keyword context and validation, and it's the one piece that could be described as a small contribution. The paper also correctly notes that Hyperscan is fast but resource-hungry, and PCRE's backtracking is a problem for untrusted input.

Where it falls down: There is no dataset, no pattern list, no model checkpoint, no code, no hardware details beyond vague classes, and no error bars. Section 4's 'average processing speed of 100 MB/s' is directly contradicted by Table 3, which gives 4.85 seconds for 100 MB (about 20 MB/s) and 530 seconds for 10 GB (about 19 MB/s). The RAM claim in the abstract (8-16 MB for RE2) is a different quantity from the end-to-end memory in Table 7 (115-190 MB), and the paper never reconciles the two. Similarly, Table 4 reports 94.8% precision and 88.7% recall for AI+Regex, but Table 5's counts at 100 MB (102 matches, 2 false positives, 2 missed) imply precision and recall around 98%. Those numbers are not consistent with a single benchmark run. The paper also cites 'Table 1.3' and 'Table 1.4' when it means Tables 1 and 2, and there are several broken sentences. These problems are exactly the kind that make a referee suspect the results are assembled rather than measured.

Who it's for: A practitioner who wants a qualitative sketch of a hybrid PII detection pipeline could get something from the architecture description. But as a research submission, it's not reproducible and its headline numbers are unreliable. It would need a complete evaluation package and a reconciliation of the tables to be worth serious review. As submitted, I'd desk-reject it with a clear list of what's missing.

Referee Report

5 major / 4 minor

Summary. The paper proposes a hybrid PII/PHI detection pipeline, CHPDA, that combines Google RE2 regular-expression scanning, Aho-Corasick exact keyword matching, AI-based named entity recognition, and validation functions such as Luhn's algorithm, followed by a proximity-based confidence-score threshold. The central claims are that RE2 is the best regex engine (10-15 ms/MB, 8-16 MB, 99.5% accuracy), Aho-Corasick is the best exact matcher (8 ms/MB), and the hybrid system attains a 91.6% F1 score with false positives below 3% and an average processing speed around 100 MB/s across file formats up to 1 TB. The paper presents benchmark tables and device measurements to support these claims.

Significance. If the headline results were reproducible, CHPDA would be a practically useful architecture for data-loss-prevention tooling, and the comparison of RE2, PCRE, Hyperscan, and exact-match algorithms is a relevant engineering question. The manuscript deserves credit for presenting the algorithm as pseudocode in Section 3.6.1, giving explicit example walkthroughs, and candidly listing limitations such as limited multilingual support and the need for pattern updates. However, the evidence supplied is not sufficient to establish the claims: there is no dataset, no pattern list, no software or hardware versioning, no repeated trials, and the numbers in the paper's own tables are mutually inconsistent. The significance is therefore conditional until a controlled, reproducible evaluation is provided.

major comments (5)

[Section 4 and Table 3] The headline speed claim of an 'average processing speed of 100 MB/s' (Section 4) is contradicted by Table 3. The reported end-to-end times for the 100-pattern regex set imply 100 MB / 4.85 s = 20.6 MB/s, 1 GB / 42.92 s = 23.3 MB/s, and 10 GB / 530.87 s = 19.7 MB/s, roughly a factor of five lower. No methodology is given to reconcile this discrepancy, so the speed pillar of the central claim is internally inconsistent.
[Section 5.4, Tables 4 and 5] The F1 claim is not supported by the paper's own detection table. Using the standard definitions and Table 5 (100 MB row: total matches 102, false positives 2, missed 2), true positives = 100 and actual positives = 102, giving precision = recall = 98.0% and F1 ≈ 98.0%, not the 94.8% / 88.7% / 91.6% reported in Table 4. The 500, 1000, and 5100 MB rows similarly yield precision and recall near 98-99% under this reading. If 'Total Patterns' is instead the ground truth, the 'Missed Matches' column is contradictory. In either case, the paper does not define what total patterns or total matches mean relative to a ground-truth corpus, so the reported F1 value is not reproducible.
[Section 4 and Section 5] The evaluation lacks an experimental protocol. No ground-truth corpus is described or cited, no regex pattern list is given, no software versions or hardware specifications beyond coarse classes are reported, and every table reports single-point numbers without repetitions, confidence intervals, or statistical tests. 'Accuracy' is never formally defined, and the claimed 'over 95% accuracy' and '<3% false positive rate' cannot be checked. Without this information, the benchmarks in Tables 1-5 are not reproducible and cannot support the comparative conclusions.
[Abstract, Table 1, Section 5.4.2, Table 7] The memory-efficiency claim for RE2 is difficult to reconcile with the system memory measurements. The abstract and Table 1 report RE2 memory usage of 8-16 MB, while Table 7 reports pipeline memory usage of 115-190 MB for regex sets of 100-200 patterns. If the former is library-only and the latter is the full CHPDA process, that distinction is never stated; if they are meant to be comparable, they contradict. The same holds for the CPU-usage percentages in Table 6, for which no measurement methodology is given.
[Section 3.6.2] The confidence-scoring example validates an SSN-like pattern '123-45-6789' with Luhn's algorithm, but Luhn's algorithm is a checksum for credit-card numbers, not for SSNs, and the example asserts without demonstration that the pattern passes. The algorithm's free parameters (alpha, Dmax, and Validation_Score in the proximity formula, plus threshold T in Section 3.7) are set by hand with no calibration or sensitivity analysis, leaving the relationship between the scoring mechanism and the reported accuracy unexplained.

minor comments (4)

[Abstract and Sections 5.1-5.2] The abstract contains a typographical spacing error ('91. 6%'), and the text refers to 'Table 1.3' and 'Table 1.4' in Sections 5.1 and 5.2, which should be Table 1 and Table 2, respectively.
[Figures 4 and 5] Figures 4 and 5 are referenced and captioned, but no actual plot images are included in the submitted text, so the claimed linear scaling trend and the match-count trend cannot be inspected.
[References] Several references are incomplete or inconsistently formatted; for example, reference [20] is only 'Xiang Wang. Introduction to Hyperscan' with no venue or year, and some citations in the introduction lack closing brackets or page ranges.
[Overall] The paper would benefit from a data-availability statement; no code, datasets, benchmark scripts, or raw measurement logs are provided, which further limits reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's benchmark claims are external measurements, not derived from its own inputs.

full rationale

CHPDA is an empirical engineering report. Its central claims (RE2 speed/accuracy, Aho-Corasick speed, AI+Regex F1=91.6%, CPU/memory usage) are presented as measurements in Tables 1-7, not as consequences of a mathematical model. No equation in the paper defines a benchmark result in terms of the result it is supposed to support. The only formula is the proximity confidence score (Section 3.6.1), Ctotal = max(0, alpha*(Dmax-d)) + Validation_Score. The constants alpha=2, Dmax=20, and Validation_Score=30 are hand-chosen in the example and are not fitted to, or derived from, the reported F1/accuracy numbers; threshold T is an adjustable user parameter, not a hidden fitted input. The paper cites prior work for RE2, Hyperscan, Aho-Corasick, and transformer NER, but none of those citations is by the present authors, and the relevant algorithmic properties (linear-time DFA matching, O(N+M) Aho-Corasick) are standard external results with independent support. Reproducibility problems (no dataset citation, no pattern set, no error bars, and an apparent contradiction between the 100 MB/s claim and Table 3's 4.85 s/100 MB times) concern experimental validity, not circularity: an unverified measurement is not a claim that reduces to its own input. I therefore find no significant circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central detection quality claims depend on four hand-set parameters (alpha, Dmax, validation score, and threshold) and on unvalidated assumptions about proximity as a confidence signal and about NER generalization. There are no fitted constants in the physical sense, but the scoring algorithm is not derived from data or from a stated optimization objective.

free parameters (4)

alpha (proximity scaling factor) = 2 (example in Section 3.6.2)
Scaling factor in Proximity_Score(d) = max(0, alpha * (Dmax - d)); user-tunable and set by hand in the worked example, with no calibration evidence.
Dmax (maximum keyword-pattern distance) = 20 (example in Section 3.6.2)
Distance threshold beyond which the proximity score becomes zero; chosen by hand in the example, with no justification from data.
Validation_Score = 30 (example in Section 3.6.2)
Additive bonus when a pattern passes a validation function such as Luhn's algorithm; chosen by hand and not derived from any objective.
Threshold T = 50 (example in Section 3.7)
Confidence cutoff for retaining detections; described as user-defined and dynamically adjustable, with no principled selection method.

assumptions (5)

standard math Aho-Corasick matching runs in O(N+M) time with a DFA built from a trie.
Invoked in Section 3.4 and cited to reference [21]; standard result, not proved in the paper.
standard math RE2 guarantees linear-time matching and safe resource use via DFA construction.
Invoked in Section 3.2 and cited to reference [19]; broadly accepted, but not proved in the paper.
domain assumption Proximity between a keyword and a detected pattern is a valid proxy for confidence that the pattern is sensitive.
This is the core heuristic of the scoring mechanism in Section 3.6; no empirical validation is provided for the assumption that closer keywords imply sensitive data.
domain assumption NER models trained on PubMed, finance, or legal corpora generalize to the private data streams being scanned.
Section 3.3 asserts these models detect PHI, names, and addresses, but no model card, training data, or evaluation is provided.
domain assumption Validation functions such as Luhn's algorithm reduce false positives without missing true positives.
Section 3.5 lists validation functions as reliable filters, but no error analysis is given to support this assumption.

invented entities (1)

CHPDA (Context-Aware Hybrid Pattern Detection Algorithm)
purpose: A named pipeline that combines RE2 regex scanning, Aho-Corasick keyword matching, validation functions, AI NER, and a confidence threshold to detect PII and PHI.
The algorithm is described in prose and pseudocode only; no implementation, benchmark harness, or external validation is provided. It is an ordinary engineering system rather than a new physical entity, but its claimed effectiveness is not independently evidenced.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Decoding Complexity: Intelligent Pattern Exploration with CHPDA (Context Aware Hybrid Pattern Detection Algorithm)." pith.science (2026). https://pith.science/paper/G4JH4CIQ

@misc{pith2026250207815,
  author       = {Pith},
  title        = {Pith review of: Decoding Complexity: Intelligent Pattern Exploration with CHPDA (Context Aware Hybrid Pattern Detection Algorithm)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G4JH4CIQ}},
  note         = {Machine review of arXiv:2502.07815}
}

read the original abstract

Detecting sensitive data such as Personally Identifiable Information (PII) and Protected Health Information (PHI) is critical for data security platforms. This study evaluates regex-based pattern matching algorithms and exact-match search techniques to optimize detection speed, accuracy, and scalability. Our benchmarking results indicate that Google RE2 provides the best balance of speed (10-15 ms/MB), memory efficiency (8-16 MB), and accuracy (99.5%) among regex engines, outperforming PCRE while maintaining broader hardware compatibility than Hyperscan. For exact matching, Aho-Corasick demonstrated superior performance (8 ms/MB) and scalability for large datasets. Performance analysis revealed that regex processing time scales linearly with dataset size and pattern complexity. A hybrid AI + Regex approach achieved the highest F1 score (91. 6%) by improving recall and minimizing false positives. Device benchmarking confirmed that our solution maintains efficient CPU and memory usage on both high-performance and mid-range systems. Despite its effectiveness, challenges remain, such as limited multilingual support and the need for regular pattern updates. Future work should focus on expanding language coverage, integrating data security and privacy management (DSPM) with data loss prevention (DLP) tools, and enhancing regulatory compliance for broader global adoption.

Figures

Figures reproduced from arXiv: 2502.07815 by the authors.

**Figure 1.** proposed System 3.2 Regex Google RE2 Pattern Recognition We chose Google RE2 as our primary regular expression detection library over alternatives such as PCRE, Oniguruma, and Boost. Regex is known for its efficiency and security in big data processing; unlike other libraries, RE2 offers linear time complexity for regular expression matching, and its SET mechanism provides an effective solution for matching multiple… view at source ↗

**Figure 2.** Regular Expression Visualization for a Numeric Pattern [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Trie Data Structure Representation Step 2: Add Failure Links Failure links are added to handle mismatches efficiently. For example, if we encounter a mismatch at h, the failure link will guide us back to the root. Similarly, failure links are created for deeper nodes in the Trie to prevent redundant comparisons. Step 3: Process the Text Now, we process the text "ushers went to her house, and his brother was there." … view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Folder Classification Time vs. Data Size for Different Regex Sets [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

**Figure 5.** Figure 5: Total Matches and Missed Matches vs. File Size [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[1]

Balancing Patient Privacy and Health Data Security: The Role of Compliance in Protected Health Information (PHI) Sharing

Amin, Md Al and Tummala, Hemanth and Shah, Rushabh and Ray, Indrajit. Balancing Patient Privacy and Health Data Security: The Role of Compliance in Protected Health Information (PHI) Sharing. arXiv preprint arXiv:2407.02766, 2024

work page Pith review arXiv 2024
[2]

Davis, James C and Michael IV , Louis G and Coghlan, Christy A and Servant, Francisco and Lee, Dongyoon Why aren’t regular expressions a lingua franca? an empirical study on the re-use and portability of regular expressions In Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Sof...

work page 2019
[3]

Vietnamese Named Entity Recognition using Token Regular Expressions and Bidirectional Inference

Le-Hong, Phuong. Vietnamese named entity recognition using token regular expressions and bidirectional inference arXiv preprint arXiv:1610.05652, 2016

work page Pith review arXiv 2016
[4]

A review on evaluation metrics for data classification evaluations

Hossin, Mohammad and Sulaiman, Md Nasir. A review on evaluation metrics for data classification evaluations. In International journal of data mining & knowledge management process, 2015

work page 2015
[5]

PAMPO: using pattern matching and pos-tagging for effective Named Entities recognition in Portuguese

Rocha, Conceição and Jorge, Alípio and Sionara, Roberta and Brito, Paula and Pimenta, Carlos and Rezende, Solange. EPAMPO: using pattern matching and pos-tagging for effective Named Entities recognition in Portuguese rXiv preprint arXiv:1612.09535, 2016

work page Pith review arXiv 2016
[6]

ECombining Regular Expressions and Supervised Algorithms for Clinical Text Classification

Flores, Christopher A and Verschae, Rodrigo. ECombining Regular Expressions and Supervised Algorithms for Clinical Text Classification. In International Conference on Intelligent Data Engineering and Automated Learning, pages 381–392, 2015

work page 2015
[7]

Integrating Regular Expressions with Neural Networks via DFA

Li, Shaobo and Liu, Qun and Jiang, Xin and Yin, Yichun and Sun, Chengjie and Liu, Bingquan and Ji, Zhenzhou and Shang, Lifeng. Integrating regular expressions with neural networks via DFA.arXiv preprint arXiv:2109.02882, 2021

work page Pith review arXiv 2021
[8]

Differential privacy and the 2020 US Census MIT Schwarzman College of Computing, 2022

Garfinkel, Simson. Differential privacy and the 2020 US Census MIT Schwarzman College of Computing, 2022

work page 2020

Show all 22 references

[9]

A Literary Review of Pattern Matching Techniques in Network Intrusion Detection

Venkatesh, Nitin and Kashikar, Pradnya. A Literary Review of Pattern Matching Techniques in Network Intrusion Detection
[10]

An investigation into performance-related issues of regular expression matching˙ Stellenbosch: Stellenbosch University, 2022

Van Litsenborgh, Pieter Steyn. An investigation into performance-related issues of regular expression matching˙ Stellenbosch: Stellenbosch University, 2022

2022
[11]

Bert: Pre-training of deep bidirectional transformers for language understanding

Kenton, Jacob Devlin Ming-Wei Chang and Toutanova, Lee Kristina. Bert: Pre-training of deep bidirectional transformers for language understanding. Proceedings of naacL-HLT, vol. 1, no. 2, 2019

2019
[12]

Honnibal, Matthew spaCy 2: Natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing No Title, 2017

2017
[13]

A survey on deep learning for named entity recognition

Li, Jing and Sun, Aixin and Han, Jianglei and Li, Chenliang. A survey on deep learning for named entity recognition. IEEE transactions on knowledge and data engineering,vol. 34, no. 1, pp. 50–70, 2020

2020
[14]

Combining Regular Expressions and Machine Learning for SQL Injection Detection in Urban Computing

Souza, Michael S and Ribeiro, Silvio ESB and Lima, Vanessa C and Cardoso, Francisco J and Gomes, Rafael L. Combining Regular Expressions and Machine Learning for SQL Injection Detection in Urban Computing. Journal of Internet Services and Applications, vol. 15, no. 1, pp. 103–...

2024
[15]

Analyzing the Efficacy of Microsoft Presidio in Identifying Social Security Numbers in Unstructured Text

Friebely, Alexander. Analyzing the Efficacy of Microsoft Presidio in Identifying Social Security Numbers in Unstructured Text. Utica University, 2022. 13 Decoding Complexity: CHPDA for Intelligent Pattern Detection

2022
[16]

Hyperscan: A fast multi-pattern regex matcher for modern {CPUs}

Wang, Xiang and Hong, Yang and Chang, Harry and Park, KyoungSoo and Langdale, Geoff and Hu, Jiayu and Zhu, Heqing. Hyperscan: A fast multi-pattern regex matcher for modern {CPUs}. 16th USENIX Symposium on Networked Systems Design and Implementation (NSDI 19), pp. 103–111, 2019

2019
[17]

RE2: a principled approach to regular expression matching

Russ Cox. RE2: a principled approach to regular expression matching. Google Research, 2010

2010
[18]

Energy and policy considerations for modern deep learning research

Strubell, Emma and Ganesh, Ananya and McCallum, Andrew. Energy and policy considerations for modern deep learning research. Proceedings of the AAAI conference on artificial intelligence, vol. 34, no. 9, pp. 13693–13696, 2020

2020
[19]

Regular expressions in big data analytics

Chowdhury, Rahul and Babu, M Rajasekhara and Mishra, Vatsal and Jain, Harshit. Regular expressions in big data analytics. 2017 International Conference on Intelligent Computing and Control (I2C2), pp. 1–10, 2017

2017
[20]

Introduction to Hyperscan

Xiang Wang. Introduction to Hyperscan
[21]

Efficient string matching: an aid to bibliographic searchCommunications of the ACM, vol

Aho, Alfred V and Corasick, Margaret J. Efficient string matching: an aid to bibliographic searchCommunications of the ACM, vol. 18, no. 6, pp. 333-340, 1975

1975
[22]

Journal Of Computer Science And Information Technologies, vol

Hasib, Saima and Motwani, Mahak and Saxena, Amit Importance of aho-corasick string matching algorithm in real world applications. Journal Of Computer Science And Information Technologies, vol. 4, pp. 467-469, 2013. 14

2013

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

Balancing Patient Privacy and Health Data Security: The Role of Compliance in Protected Health Information (PHI) Sharing

Amin, Md Al and Tummala, Hemanth and Shah, Rushabh and Ray, Indrajit. Balancing Patient Privacy and Health Data Security: The Role of Compliance in Protected Health Information (PHI) Sharing. arXiv preprint arXiv:2407.02766, 2024

work page Pith review arXiv 2024

[2] [2]

Davis, James C and Michael IV , Louis G and Coghlan, Christy A and Servant, Francisco and Lee, Dongyoon Why aren’t regular expressions a lingua franca? an empirical study on the re-use and portability of regular expressions In Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Sof...

work page 2019

[3] [3]

Vietnamese Named Entity Recognition using Token Regular Expressions and Bidirectional Inference

Le-Hong, Phuong. Vietnamese named entity recognition using token regular expressions and bidirectional inference arXiv preprint arXiv:1610.05652, 2016

work page Pith review arXiv 2016

[4] [4]

A review on evaluation metrics for data classification evaluations

Hossin, Mohammad and Sulaiman, Md Nasir. A review on evaluation metrics for data classification evaluations. In International journal of data mining & knowledge management process, 2015

work page 2015

[5] [5]

PAMPO: using pattern matching and pos-tagging for effective Named Entities recognition in Portuguese

Rocha, Conceição and Jorge, Alípio and Sionara, Roberta and Brito, Paula and Pimenta, Carlos and Rezende, Solange. EPAMPO: using pattern matching and pos-tagging for effective Named Entities recognition in Portuguese rXiv preprint arXiv:1612.09535, 2016

work page Pith review arXiv 2016

[6] [6]

ECombining Regular Expressions and Supervised Algorithms for Clinical Text Classification

Flores, Christopher A and Verschae, Rodrigo. ECombining Regular Expressions and Supervised Algorithms for Clinical Text Classification. In International Conference on Intelligent Data Engineering and Automated Learning, pages 381–392, 2015

work page 2015

[7] [7]

Integrating Regular Expressions with Neural Networks via DFA

Li, Shaobo and Liu, Qun and Jiang, Xin and Yin, Yichun and Sun, Chengjie and Liu, Bingquan and Ji, Zhenzhou and Shang, Lifeng. Integrating regular expressions with neural networks via DFA.arXiv preprint arXiv:2109.02882, 2021

work page Pith review arXiv 2021

[8] [8]

Differential privacy and the 2020 US Census MIT Schwarzman College of Computing, 2022

Garfinkel, Simson. Differential privacy and the 2020 US Census MIT Schwarzman College of Computing, 2022

work page 2020