ScholarWorks@숙명여자대학교

초록

Crowdfunding has recently been recognized as a prime example of the process economy. This funding method is becoming an important platform that replaces traditional funding approaches. However, existing research on crowdfunding success factors has focused mainly on quantitative variables such as goal amounts and fulfillment rates. The potential importance of unstructured data, particularly project description text, has been relatively overlooked. To address these limitations, this study empirically investigates how project description text impacts crowdfunding success. We employ LDA topic modeling and machine learning techniques using data from Tumblbug, a leading Korean sponsored crowdfunding platform. The analysis identified 10 major topics, including Traditional Culture, Character Goods, and Festivals & Seasonal Items. Among various machine learning algorithms, Gradient Boosting performed the best in all evaluation metrics. Adding LDA topic variables to the basic variables improved prediction performance in all categories. For example, the study observed improvements of 6% in Character Goods and 4% in publishing categories. These results suggest that crowdfunding functions as an alternative distribution channel. It provides differentiated experiences through artistic and cultural values. This study presents a novel methodology that combines text mining and machine learning. This approach enables researchers to derive customized strategies for each category, offering both academic and practical implications.

키워드