Default Identification of Scientific Crowdfunding Projects: A Hesitant Fuzzy Deep Learning Approach

Abstract

To promote and achieve innovations, crowdfunding projects have emerged as a novel tool for attracting financial support for scientific research. However, before initiating a crowdfunding campaign, only the project's presentation and contractual terms can be disclosed, and then reasonable judgment and evaluation are important to make an informed investment decision regarding the crowdfunding project. To do this, this article introduces the probabilistic hesitant fuzzy set (PHFS) into the evaluation process and proposes the probabilistic hesitant recursive neural network (PH-RNN). The PH-RNN represents an innovative approach within the realm of hesitant deep learning, employing PHFS to comprehensively articulate subjective cognitive insights. Subsequently, diverse descriptions can be meticulously considered in the default identification process, thereby facilitating the precise identification of potential defaults for scientific crowdfunding projects. In this modeling process, a data processing layer for the PH-RNN is developed, which can effectively integrate the probabilistic hesitant fuzzy information. Moreover, a default recognition algorithm is further designed to address the complexities of deep learning in conjunction with the PHFS. Conclusively, an empirical study conducted on the scientific crowdfunding platform “Experiment” is presented to demonstrate the proposed models and process.