Predicting Startup Success Through a New Graph Convolutional Neural Network Approach

Venture capital (VC) is a high-risk, high-return capital injection for emerging or small and medium-sized enterprises with growth potential. Accurately predicting the success of startup is critical for venture capital institutions. Previous studies have used data-driven approaches to explore the many factors that influence startup success, but have often failed to delve deeper into the heterogeneous topological information in investment relationships. In fact, graph convolutional neural networks have been shown to be effective for binary classification problems. In this paper, to model the relational embedding of co-investment networks and investment relationship networks, a novel graph convolutional neural network approach, named VC-HGCN, is employed to effectively capture the complex interactions in investment networks by integrating the learned node attributes and network topology information. The results show that VC-HGCN performs better in predicting startup success compared to the benchmark model, and also demonstrates the value of using the two investment networks in combination. Our study improves the accuracy of startup success prediction and provides a powerful tool for venture capitalists to find high-potential startups.