A feature interaction learning approach for crowdfunding project recommendation

Abstract Crowdfunding is an emerging internet platform that provides financial support for people in need. With the development of crowdfunding platforms, the number of projects released on these platforms is increasing, and thus it is challenging for lenders to find suitable crowdfunding projects quickly. A personalized recommender system is helpful for solving this problem. To this end, a crowdfunding project recommendation approach is proposed in this work for predicting how likely a lender is to fund a project. Specifically, given the fact that the lenders consider not only their interests but also the benefits they can obtain when funding, we first design a module that predicts the success rate of the projects. Then, a feature interaction learning model based on deep learning, called the crowdfunding feature interaction learning model, is proposed. It integrates all features, automatically recognizes the importance of features, and learns the feature interaction. This allows us to identify combination of useful features more accurately and provide effective predictions. Extensive experiments are conducted on a dataset collected from a real-world crowdfunding platform, and the results show that our approach has a 4.57% improvement on AUC (area under the curve) compare with the state-of-the-art methods.