Fine-grained Feature Fusion Framework for Online Crowdfunding Success Prediction

Abstract

Crowdfunding provides a new solution for individuals and companies to overcome financial hardships. However, how to improve the success rate of crowdfunding remains a challenge for project initiators. Pre-launch crowdfunding success prediction allows the initiator to understand the likelihood of crowdfunding success and then adjust the project information based on the result. Previous deep learning-based pre-launch crowdfunding success prediction models mainly focused on improving model performances by applying cutting-edged AI models or algorithms. These methods ignored the fine-grained features contained in projects and platforms that cannot be recognized by pre-trained encoders. In this study, we use speech act theory to recognize linguistic patterns in project descriptions. We will also apply contestable market theory to capture the fine-grained regional features as well as competition intensity on the platform and use social exchange theory to add key information that may affect donors’ decisions in impact letters to the framework. The experiment results demonstrate the effectiveness of using the fine-grained features in pre-launch crowdfunding success prediction.