Who said what: Mining semantic features for success prediction in reward-based crowdfunding

Abstract

• Novel semantic features for success prediction in reward-based crowdfunding. • Proposed features rooted in information asymmetry and herding behavior theories. • Informational comments and inclinational comments of different actor roles. • Empirical evaluation demonstrates the utility of the proposed framework. As the reward-based crowdfunding market is growing rapidly, it becomes increasingly important for stakeholders to effectively predict fundraising outcomes (i.e., success or failure). More and more participants share and discuss facts and opinions about projects by posting comments, which can influence investors’ funding decisions. Previous studies have mainly focused on quantity, sentiment, and linguistic features of comments, largely overlooking the value of semantic features, in predicting fundraising success. Rooted in information asymmetry and herding behavior theories, we posit that discovering semantic signals from comments and distinguishing actor roles may benefit fundraising success prediction. We propose a framework with novel latent semantic features of comments. Empirical evaluation using data from a prominent platform demonstrates the utility of the framework and reveals interesting patterns in the dynamic predictive effects of semantic features for different actor roles.