A Structure Alignment Deep Graph Model for Mashup Recommendation

Eduardo Lima, Xumin Liu

Abstract: In this paper, we propose a novel approach to recommend services for a given mashup development task. We model service data as a heterogeneous service graph which includes multiple types of nodes and edges to capture rich information extracted from the data. We extend the design of the graph convolutional networks to learn optimal graph embeddings based on a novel structure alignment framework leveraging the latent heterogeneous graph structural features. We then design a ranking mechanism to recommend those services so that their links to the mashup can best fit the latent graph structural features. Both the embedding learning and ranking process make the use of meta-paths to incorporate prior domain knowledge into recommendation. A comprehensive experimental study is conducted on a real-world data set and the result indicates that our approach can significantly outperform the existing solutions.