DBLab
School of Computer Science and Engineering, Soongsil University
DBLab studies how data is stored, managed, and analyzed. Starting from flash-memory storage systems, our work has grown to include digital forensics, voice-phishing detection, and rights-metadata extraction for openly licensed works.
Research
All research →
NLP & Rights Metadata
2024 – present
A NER + LLM “division of labor” pipeline that extracts rights metadata from Korean openly licensed works, with per-field provenance.
Voice Phishing Detection
2021 – present
KorCCVi, the first labeled Korean voice-phishing corpus, and models from KoBERT to multimodal audio–text and privacy-preserving real-time detection.
Digital Forensics
2010 – 2020
Recovery of deleted and wiped files on NAND flash and SSDs, YAFFS2 metadata, and SQLite deleted-record and messenger forensics.
Flash Storage & FTL
2004 – present
Log-buffer flash translation layers (FAST, LSTAFF*), hybrid address mapping, flash-aware B-tree indexes, and crash recovery.
Recent News
- Two papers accepted at IC-EEECS 2026 — the NER–LLM consolidation pipeline for metadata extraction from Korean public-domain rights documents, and the KLUE-BERT B-I-O training-data composition analysis.
- Research impact: the FAST flash translation layer paper (ACM TECS 2007) has passed 700 citations.
- Revised manuscript on lightweight on-device vishing detection submitted to IEEE Access (revise-and-resubmit).
- Multimodal audio–text Korean vishing detection presented at KCC 2025, with four student co-authors.
- The lab joins the Muhayu-led MCST/KOCCA R&D consortium building copyright-metadata AI for openly licensed works (with HM Company; demand organizations: Korea Copyright Commission and KCISA).
- IEEE Access publishes our study on enhancing voice-phishing detection with multilingual back-translation and SMOTE.