Success Prediction of Crowdfunding Campaigns with Project Network: A Machine Learning Approach


Chao Zhong
Wei Xu
Wei Du


In the last decade, crowdfunding has emerged as a new form of Internet finance, providing founders with a channel through which they can raise funds from the public. Prior studies have mainly investigated two types of crowdfunding success predictors: conventional numerical features (e.g., project goal, duration, number of rewards, number of comments and the presence of a video) and features extracted from textual description and project images. In comparison, few studies have examined the effect of interrelations among projects on crowdfunding performance. For example, a founder can learn from historically invested projects when launching one’s own project. In this study, we extend the previous understanding by introducing the concept of “project network,” which can be constructed by extracting founders’ activities on crowdfunding platforms. Network-based features are extracted from the project network through the Node2vec method. Experimental results show that models with network-based features outperform those without network-based features. Furthermore, the dense dataset with densely connected projects achieves better prediction performance than the original one, further validating the role of a project network in success prediction. Another implication is that a small proportion of connected projects could help predict project success to avoid a high calculation cost.

Key Word: 

Published Date: 

April, 2022

Full File: