Software & Data

Software & Data 소프트웨어 & 데이터

Datasets and code the community actually uses — versioned, licensed, and citable.

Dataset

KorCCVi — Korean Call Content Vishing

The first labeled Korean voice-phishing transcript dataset, grown across versions from v1.3 to v3; the multimodal line adds a call-audio corpus of 1,417 calls (706 vishing / 711 non-vishing) with 39,429 utterance segments — 66.9 hours at 16 kHz mono.

v1.3 ~6,005 rowsv2.1 ~20,857 rowsv3 ~18,646 rowsaudio 1,417 calls / 66.9 h

GitHub

Dataset

Augmented Vishing Dataset

KorCCVi with multilingual back-translation and SMOTE augmentations — 2,927 original samples (695 vishing / 2,232 non-vishing) plus ~3,502 augmented samples, released on IEEE DataPort in November 2024.

DOI 10.21227/163c-0542CC BY-NC-SA 4.0216 MB

IEEE DataPort

Code

Korean_Voice_Phishing_Detection

Open code for the vishing paper series — ML/DL text classifiers, KoBERT, attention CNN-BiLSTM, federated learning, and multimodal audio–text fusion.

public repo24 starsPython/PyTorch

GitHub

KorCCVi

KorCCVi Versions 데이터셋 버전

KorCCVi has grown through successive releases — size differences between versions are deliberate releases, not conflicts.

KorCCVi dataset versions, contents and sizes
Version	Type	Size	Notes
KorCCViD v1.3	Text	~6,005 rows	Labeled call transcripts — the original release of the KorCCVi line (introduced at KCC 2021).
KorCCVi v2 / v2.1	Text	~20,857 rows	Expanded transcript corpus.
KorCCVi v3	Text	~18,646 rows	Latest text release.
Audio corpus	Audio	1,417 calls · 39,429 segments · 66.9 h	16 kHz mono — 706 vishing / 711 non-vishing calls (multimodal research line).
Augmented (IEEE DataPort)	Text	2,927 original + ~3,502 augmented	Multilingual back-translation + SMOTE · CC BY-NC-SA 4.0 · released Nov 2024.

Papers behind these datasets