Hybrid Model of Crowdfunding Prediction with User Analysis and Neural Networks Algorithm

Extending recent empirical work on testing what factors affect crowdfunding performance, we construct a new hybrid model to evaluate crowdfunding performance by analyzing user behavior with neural networks algorithm. Project signals observation and user preferences for uncertainty are modeled as the behavior-related factors. By considering user behavior, the sparse optimization is designed with the neural networks algorithm to get the optimal feature subset. To validate the proposed model, we do the experiments based on the data of a crowdfunding platform, in which the data-set including 151,752 crowdfunding projects. The experiments results show that, on the one hand, the recommendation order of the candidate projects with the proposed model is consistent with the order identified from the actual statistical data. On the other hand, the proposed model has better predictive power in predicting crowdfunding performance.