?
Мультимодальное прогнозирование успешности краудфандинговых проектов: сигнальная роль текстового и визуального контента
Crowdfunding is a relatively new method of project financing that involves raising funds from a large number of non-professional investors. Despite the rapid growth of this sector, only about 41.98% of crowdfunding campaigns are successful, which highlights the need to identify factors that increase the likelihood of successful fundraising. This study aims to address the question of which textual and visual characteristics of projects most significantly determine their success and to what extent their inclusion improves forecasting accuracy. The study develops a multimodal model for predicting the likelihood of fundraising on the Kickstarter platform. The study hypothesizes that the inclusion of textual (H1) and visual (H2) characteristics significantly improves the predictive power of the models, and their joint integration (H3) ensures the best forecast quality by incorporating additional information. The empirical analysis relies on a dataset of 89,557 projects launched on Kickstarter between 2010 and 2024.
In addition to standard project characteristics commonly used in studies of crowdfunding success (funding goal, campaign duration, category, country, and launch year), the analysis incorporates an extended set of textual and visual factors. Textual features are extracted using natural language processing techniques, including sentiment analysis, readability, and logicality measures applied to project descriptions and risk disclosures. Visual features are obtained using a pretrained ResNet50 neural network, followed by dimensionality reduction via principal component analysis. For predictive purposes, logistic regression, decision trees, ensemble methods (Random Forest, XGBoost, LightGBM), and a multilayer perceptron are employed.
The results demonstrate that the inclusion of textual and visual factors significantly improves predictive performance. The best-performing model (LightGBM) achieves a ROC AUC of approximately 0.82. The findings indicate that funding goals, campaign duration, textual characteristics, and visual content exert the strongest influence on the probability of success. Based on these results, the study concludes that informational signals play a significant role in crowdfunding and offers practical recommendations for improving fundraising strategies on crowdfunding platforms.