Functional Complementarity-Aware Representation Learning for Cold-Start Recommendation in Mashup Creation

Cloud APIs serve as critical facilitators for data exchange, capability replication, and service delivery, having become integral to the digital infrastructure that underpins modern service-oriented software development and operations. However, the rapid proliferation of available cloud APIs poses considerable challenges for developers in selecting the most appropriate cloud APIs to fulfill their development requirements. Current methods primarily rely on historical interaction data, which are particularly inadequate in cold-start scenarios where such data is scarce. Furthermore, these approaches often neglect the complementary relationships that inherently exist between mashups and cloud APIs. To this end, we propose a functional complementarity-aware representation learning (FCRL) model for mitigating cold-start in cloud API recommendation. FCRL has a three-fold main idea: 1) constructing a functional complementary graph by examining tag co-occurrence patterns between Mashups and cloud APIs; 2) exploiting a graph convolutional network to derive representations that encapsulate functional complementarity for both mashups and cloud APIs; and 3) utilizing an attention mechanism to dynamically integrate representations of functional complementarity, collaborative interaction, and content information, thereby producing high-quality feature embeddings for cold-start mashups and cloud APIs. Empirical evaluations demonstrate that FCRL significantly outperforms existing methods in both sparse and cold-start scenarios. Furthermore, the incorporation of attention weights within FCRL facilitates the identification of the primary types of information that influence predictions, thereby enhancing the understandability of FCRL. Our code is available at https://github.com/whitewolf666/FCRL.