This study aims to empirically identify the structural patterns of social support exchanged by cancer survivors in online communities and their unmet socio-economic needs during labor market reintegration, by analyzing YouTube vlog comments as a form of naturally occurring unstructured big data. As Korea's five-year cancer survival rate has reached 72.1%, cancer survivors increasingly face challenges of employment discontinuity and job instability, yet existing social safety nets fail to adequately address these needs. We collected 164,850 raw comments from 150 cancer survivor vlog videos posted between January 2020 and December 2024, employing a dual sourcing strategy combining YouTube Data API v3 and Selenium web drivers. Data collection was completed by February 2025, and preprocessing and analysis were finalized by November 2025. Following a four-stage preprocessing pipeline grounded in pilot-tested thresholds, 148,708 valid comments were analyzed using TF-IDF, Latent Dirichlet Allocation (LDA) topic modeling, binomial logistic regression, and KOSAC -based sentiment analysis. LDA topic modeling (K=5, coherence score Cv=0.52) revealed that ‘emotional solidarity and support’ constituted the largest latent topic (29.6%), demonstrating that cancer survivors prioritize emotional bonding over medical information exchange in online communities. Logistic regression confirmed that emotional support vocabulary significantly predicted empathic responses measured by comment ‘like’ counts (beta=1.847, p<.001), while medical information terms showed no statistical significance. Sentiment analysis revealed dominant positive sentiment (65%) with coexisting ambivalence structures in uncertainty-related topics. Theoretically, based on the observed complex support behaviors specific to vocational reintegration contexts, we propose ‘vocational rehabilitation support’ as a new subtype extending House's (1981) social support typology. From an industrial and systems engineering perspective, the findings provide an empirical foundation for designing return-to-work (RTW) process models and job redesign decision frameworks within human resource management (HRM) systems. Practically, we derive evidence-based policy recommendations including peer support programs, flexible work arrangements, and an incentive-penalty system modeled on Germany’s Schwerbehinderten-Ausgleichsabgabe.

1. 서 론
2. 이론적 배경
    2.1 사회적 지지 이론과 직업 재활 지지
    2.2 보상적 행동과 직업 가치관 변화
    2.3 LDA 토픽 모델링의 방법론적 타당성
    2.4 연구가설
3. 연구 방법론
    3.1 연구 설계 및 데이터 수집
    3.2 데이터 전처리 파이프라인
    3.3 분석 기법
4. 연구 결과
    4.1 기술통계 및 TF-IDF 핵심 키워드
    4.2 LDA 토픽 모델링 결과
    4.3 로지스틱 회귀분석 결과
    4.4 감성 분석 및 양가감정 교차 분석
5. 논 의
    5.1 학술적 함의
    5.2 실무적․정책적 제언
6. 결 론
References
Appendix. Top 50 TF-IDF Keywords and Category Classification

• Insoo Kim(Graduate School of the Management of Technology, Jeonbuk National University) | 김인수 (전북대학교 일반대학원 융합기술경영학과(MOT))
• Joonsoo Bae(Graduate School of the Management of Technology, Jeonbuk National University) | 배준수 (전북대학교 일반대학원 융합기술경영학과(MOT)) Corresponding author
• Yeonjoo Chae(Graduate School of the Management of Technology, Jeonbuk National University) | 채연주 (전북대학교 일반대학원 융합기술경영학과(MOT))