Predicting reward-based crowdfunding success with multimodal data: A theory-guided framework

Abstract

There is a growing need to investigate the impact of multimodal data, which are becoming increasingly prevalent on crowdfunding platforms, on prediction of fundraising outcomes. However, a prediction framework drawing upon rational theoretical foundations to leverage multimodal data in crowdfunding is still lacking. Guided by relevant theories, we explore the ideational, interpersonal, and textual metafunctions of multimodal data geared toward fundraising success prediction. Our empirical evaluation demonstrates superior predictive utilities of various metafunction-based multimodal features over purely data-driven ones. Our results also reveal that the multiple data modalities interact complementarily and synergistically to improve the prediction performance. Specifically, combining metafunctions improved prediction performance by 2–15 % over a single metafunction, while multimodality outperformed single data modality by 7–18 % within each metafunction.