MENTHOS-spam

English

Model Description

MENTHOS-Spam is a binary spam classifier fine-tuned from answerdotai/ModernBERT-base for phishing/spam message detection. It uses a maximum sequence length of 256.

Intended Use

• Detect spam/phishing text in email/SMS-like content.
• Labels: 0 = ham, 1 = spam.

Not intended for legal/forensic final decisions without human review.

Training Data

• Trained on the MENTHOS spam dataset.

Benchmark Results

model	samples	accuracy	precision	recall	f1	roc_auc	p50 latency (ms)	throughput (samples/s)
MENTHOS-spam	13128	0.991392	0.994178	0.988574	0.991368	0.999480	6.1142	163.37

Reference baseline (Morpheus ONNX):

baseline model	accuracy	f1	p50 latency (ms)	throughput (samples/s)
phishing-bert-20230517.onnx	0.542581	0.675142	63.2982	15.57

Benchmark Plots

Limitations

• Dataset composition is specific to downloaded sources and preprocessing pipeline.
• Performance may degrade on different domains/languages.
• Threshold and calibration may need adaptation for production.

Citation

@misc{borovic_li-dobnik_kranjec_ferme_2026,
  title        = {MENTHOS-spam},
  author       = {Borovic, Li Dobnik, Kranjec, Ferme},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/LHRS-UM-FERI/MENTHOS-spam}}
}

---

Slovenščina

Opis modela

MENTHOS-Spam je binarni klasifikator nezaželene pošte (spam), naučen iz osnovnega modela answerdotai/ModernBERT-base.

Kartica povzema družino modelov: Uporablja maksimalno dolžino zaporedja 256.

Namen uporabe

• Zaznavanje spam/phishing besedil v vsebinah tipa e-pošta/SMS.
• Oznake: 0 = ham, 1 = spam.

Model ni namenjen samostojnim pravnim ali forenzičnim odločitvam brez človeškega pregleda.

Učni podatki

• Učenje je potekalo na MENTHOS spam datasetu.

Rezultati benchmarka

Benchmark results for the MENTHOS evaluation set.

model	vzorcev	accuracy	precision	recall	f1	roc_auc	p50 latenca (ms)	prepustnost (vzorcev/s)
MENTHOS-spam	13128	0.991392	0.994178	0.988574	0.991368	0.999480	6.1142	163.37

Referenčni baseline (Morpheus ONNX):

baseline model	accuracy	f1	p50 latenca (ms)	prepustnost (vzorcev/s)
phishing-bert-20230517.onnx	0.542581	0.675142	63.2982	15.57

Grafi benchmarka

Omejitve

• Sestava podatkov je vezana na uporabljene vire in obdelavo.
• Na drugih domenah/jezikih je lahko uspešnost slabša.
• Za produkcijo je priporočena dodatna kalibracija praga.

Citiranje

@misc{borovic_li-dobnik_kranjec_ferme_2026,
  title        = {MENTHOS-spam},
  author       = {Borovic, Li Dobnik, Kranjec, Ferme},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/LHRS-UM-FERI/MENTHOS-spam}}
}