GSL-Mash: Enhancing Mashup Creation Service Recommendations Through Graph Structure Learning

Sihao Liu, Mingyi Liu, Tianyu Jiang, Shuang Yu, Hanchuan Xu, Zhongjie Wang

Abstract: The proliferation of Web APIs has facilitated the creation of numerous software applications through the integration of diverse services, commonly referred to as mashups. However, the growing complexity and number of available Web APIs pose significant challenges in API services selection. Current service recommendation models, predominantly based on Graph Neural Networks (GNNs), often underperform due to the simplistic and overly complex APIs co-occurrence graphs they utilize, which impede both efficiency and performance. This paper introduces a novel model, GSL-Mash, which incorporates graph structure learning (GSL) to optimize graph data in service recommendations. By refining the graph structure to retain only pertinent connections, our model significantly reduces unnecessary complexity and noise, enhancing both the efficacy and accuracy of service recommendations. We validate GSL-Mash using real-world datasets from ProgrammableWeb, where it outperforms established baselines with up to \(45.39\%\) improvement on NDCG@10 metric. Additionally, we contribute to the academic and development communities by making our implementation publicly available. This study not only advances the technology of service recommendation systems but also sets a foundational approach for future research in optimizing graph-based service recommendation models.