더 나은 사회를 위한 머신러닝: 크라우드펀딩 기부 캠페인 성공 요인 분석

초록

This study combines natural language processing with econometric modeling to examine how crowdsourcing donation campaign performance varies with emotional appeals in campaign descriptions. Using all donation projects posted on a specific date, we extracted 28 discrete emotion scores and reduced them via PCA into three emotion dimensions: Activation vs. Prosocial Restraint (PC1), Social Warmth vs. Cognitive Ambiguity (PC2), and Positive Engagement vs. Negative Avoidance (PC3). PC2—loading positively on admiration, gratitude, and love, and negatively on surprise, curiosity, and confusion—showed a statistically significant positive effect on campaign progress rates. Models incorporating beneficiary type and campaign category as moderators revealed significant main effects but no significant interactions with emotion dimensions: animal-protection campaigns exceeded the baseline (children and adolescents) while elderly and vulnerable-family-women beneficiaries fell below it; education, arts, awareness/policy, and housing categories all showed lower progress rates than livelihood and basic-needs support. Moving beyond single or dichotomous emotional appeal analyses to a multidimensional 28-emotion structure, this study demonstrates that strengthening Social Warmth expressions universally enhances funding performance across beneficiary types and campaign categories, offering meaningful implications for campaign design and strategic communication.

키워드