Research On Crowdfunding Project Prediction And Personalized Recommendation Method Based On Feature Interaction Learning

With the popularization of the Internet and the sharing economy,crowdfunding,as an emerging Internet financial product,attracts extensive attention and gradually grows into an efficient business that promotes product and service innovation.However,while crowdfunding is developing at a high speed,it also exposes many deficiencies,such as information asymmetry,imperfect credit system,lack of efficient exit mechanism,potential default risk and other pain points in the Internet financial industry,resulting in uneven projects released by the crowdfunding platform,some of which have high default risk.Thus,investors are trapped in the problem of how to select investment objects under many projects.Based on the data-driven model,this dissertation focuses on the two closely related core issues of crowdfunding project success rate prediction and personalized recommendation,studies the crowdfunding project success rate prediction method,and constructs the personalized recommendation model of crowdfunding project on this basis.The main innovative work of this dissertation is as follows:(1)In order to solve the problem of crowdfunding project success rate prediction,this dissertation proposes a project success rate prediction model based on feature interactive learning.The model uses embedding operation to transform features to generate low dimensional dense vectors,uses feature interactive learning to generate combined features to represent the association of features,and adjusts the loss function to solve the impact of data imbalance on the model,so as to achieve efficient and accurate prediction.(2)In order to solve the problem of recommending crowdfunding projects for investors,this dissertation comprehensively considers the interests and interests of investors,and proposes a project recommendation model based on feature interactive learning.The model uses the project success rate obtained in the project success rate prediction stage to integrate the characteristics of the project and investors.The model automatically learns the importance of features,and uses the parallel structure to learn the interaction of low-order and high-order features at the same time,so that the model can identify useful combined features more accurately and provide effective prediction.(3)This dissertation makes extensive experiments on the real crowdfunding data set,the experimental results show that the proposed methods are significantly better than the baselines in each metric.