Reproducibility — ICTSS 2026

Testing Domain-Specific Email and Phone Detection for GDPR Compliance in Cloud Food Data Lakes.

This directory contains the material to reproduce the evaluation (Section 5 of the paper).

Contents

File	Description
pii-test-corpus.csv	Annotated corpus (500 samples: 300 negative / 200 positive)
PersonalDataCheckService.java	Detection service (3-layer pipeline)
PersonalDataException.java, FileProcessingException.java, PersonalDataWarningConfirmationRequiredException.java	Exceptions used by the service
keywords.txt	51 privacy keywords (Layer 3)
PersonalDataCheckServiceTest.java	15 unit tests
PersonalDataCheckServiceMetamorphicTest.java	207 generated metamorphic tests (MR1–MR4)
PersonalDataCheckServiceCorpusTest.java	Corpus precision/recall assertion
gen_corpus_500.py	Reproducible corpus generator (fixed seed)
eval_presidio.py	Microsoft Presidio v2.2 comparison baseline
requirements.txt	Python dependencies for the Presidio baseline

Reproduce the detector results (Java)

The Java sources here are extracted from a Spring Boot / Maven module. To run the tests, place them in a Maven project providing spring-boot-starter-test, Google libphonenumber, and Lombok, with keywords.txt on the main classpath and pii-test-corpus.csv on the test classpath (src/main/resources and src/test/resources). Then:

# Precision/recall on the corpus: expect FP=0 on 300 negatives, in-scope recall 99.3%
./mvnw test -Dtest=PersonalDataCheckServiceCorpusTest

# Metamorphic relations: expect 207 generated tests, all passing
./mvnw test -Dtest=PersonalDataCheckServiceMetamorphicTest

# Unit tests
./mvnw test -Dtest=PersonalDataCheckServiceTest

Reproduce the Presidio comparison (Python)

Python 3.10–3.13 (Presidio 2.2.362 requires <3.14):

python -m venv .venv
. .venv/bin/activate        # Windows: .venv\Scripts\Activate.ps1
pip install -r requirements.txt
python eval_presidio.py     # expect 28 false positives on food data

Regenerate the corpus (optional)

python gen_corpus_500.py    # deterministic; idempotent once the corpus has 500 rows

Expected headline numbers

System	FP (food data)	In-scope precision	In-scope recall
Ours	0 / 300	1.000 (95% Wilson CI [0.99, 1.0])	0.993
Presidio v2.2.362	28 / 300	0.833	0.993

Fisher's exact test on the false-positive difference: p < 1e-8.