You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

119 Call Transcript Synthetic Labels

한국 119 응급 신고 통화를 바탕으로 만든 합성(synthetic) 텍스트에, Claude Haiku 4.5로 구조화 라벨링(연령/성별/의식/활력징후/증상/KTAS 등급 등)을 붙인 데이터셋입니다.

This dataset contains synthetic Korean 119 (emergency call) transcripts, labeled with structured fields (age, sex, consciousness, vital signs, symptoms, KTAS triage level, etc.) using Claude Haiku 4.5.

데이터 구성 / Structure

  • train.jsonl: 7,196 rows
  • val.jsonl: 1,794 rows

전체 8,990건은 모두 합성(synthetic) 통화이며, 실제 통화 STT 기반 데이터는 포함하지 않습니다. train/val은 8:2로, Pre-KTAS 등급과 통화 종류별로 층화해 나눴습니다. 경계 사례(edge) 20건은 train에 포함됩니다. meta.data_v3_id는 내부 원본 id입니다.

All 8,990 records are synthetic; no real call transcripts are included. Split is 80/20, stratified by Pre-KTAS level and call type; the 20 edge cases are in train.

각 라인은 하나의 JSON 레코드이며 다음 필드를 포함합니다:

  • id: 레코드 고유 ID
  • split: train 또는 val
  • text: 합성된 119 통화 텍스트
  • meta: 시나리오 메타데이터(원본 소스 ID, 검증 이슈 등)
  • v2: Claude Haiku 4.5로 라벨링한 구조화 필드 (call_type, ktas_level, ktas_evidence, chief_complaint, suspected_diagnosis, vitals, consciousness, symptoms, onset, incidents, disease_category, injuries, treatments, age, sex, medications, notes)

Each line is a single JSON record with the following fields:

  • id: unique record identifier
  • split: train or val
  • text: the synthetic 119 call transcript
  • meta: scenario metadata (original source id, verifier notes, etc.)
  • v2: structured fields labeled by Claude Haiku 4.5 (call_type, ktas_level, ktas_evidence, chief_complaint, suspected_diagnosis, vitals, consciousness, symptoms, onset, incidents, disease_category, injuries, treatments, age, sex, medications, notes)

라벨링 방법 / Labeling

  • 라벨은 Claude Haiku 4.5를 사용해 생성했습니다.

  • 텍스트 자체도 합성 데이터(synthetic)이며, 실제 개인의 신고 내용이 아닙니다.

  • Labels were generated using Claude Haiku 4.5.

  • The call transcripts themselves are synthetic — they are not real emergency call recordings or transcriptions, and do not correspond to any real individual's report.

  • Each record covers a single simulated 119 emergency call scenario, labeled for downstream tasks such as KTAS (Korean Triage and Acuity Scale) level prediction, chief-complaint extraction, and vital-sign / symptom structuring.

라이선스 / License

MIT

Downloads last month
51