Multi-Relationship and Multi-Attribute Fusion Based on Graph Neural Network for API Recommendation

With the development of SOA (Service Oriented Architecture), more and more APIs are published on the Web, locating the perfect API for Mashup creation can be a challenging task. Deep learning is one of the most popular recommendation technologies due to its advantages in automatic feature extraction based on training data. However, existing deep learning-based API recommendation methods almost do not consider the multiple relationships among APIs with respect to Mashup functions, description features, categories, and multiple attributes of APIs themselves, resulting in severe “long-tail” problems and poor recommendation accuracy. In order to solve these problems, we propose a method named MRMA_GNN (Multi-Relationship and Multi-Attribute fusion based on Graph Neural Network). Concretely, three API relationship graphs are created and fused them into a multi-relationship graph matrix through average pooling Firstly. Then, API feature vectors are produced by combining the attributes vectors and description vector. Finally, the API multi-relationship graph matrix and the API feature matrix consists of all API feature vectors are input into a GNN model for API combination prediction, and thus creates API recommendation. The experimental results show the superiority of this method.