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.

Subtitle Summary & Keyword — Test set (YouTube)

방송 유튜브 자막 기반 누적 요약 + 검색 키워드 태스크의 평가용 테스트셋입니다. jungsanghyun/subtitle-summary-dataset의 train/validation과 겹치지 않는 별도 방송으로, 동일 파이프라인(Qwen3-235B, vLLM, 자막-only)으로 생성했습니다.

규모

split 회차 레코드(5분)
test 114 519

약 25시간 · 요약 중앙값 59자 · 도메인: Entertainment 236 · News & Politics 103 · Pets & Animals 87 · Travel & Events 34 · Education 31 · Music 28

스키마

program_name, last_summary, 5min_script(입력) → current_summary, current_5min_keyword(출력) + keyword_candidates + 메타. step 0-based.

from datasets import load_dataset
ds = load_dataset("jungsanghyun/subtitle-summary-testset")   # ds["test"]

학습용 messages: jungsanghyun/subtitle-summary-testset-sft.

라이선스

자막 원 저작권은 각 방송사/채널. 연구·비상업(cc-by-nc-4.0).

Downloads last month
20