Integrating Functional and Structural Semantics for Web API Recommendation via Multi Meta-path Aggregation

With the increasing popularity of Internet services, developers can reuse or mix these services to create new applications (such as mashups) to improve development efficiency. In previous studies, several web API recommendation methods have been proposed, some of which rely on mashup demand information to extract necessary functional features. These methods often incorporate Quality of Service (QoS) assessments of network services to enhance recommendations. Others utilize graph neural networks to extract implicit structural semantics from the web API network and apply them to build downstream tasks. Despite the significant contributions of these methods in advancing web API recommendations, the challenge lies in how to better extract structural information in highly sparse conditions and combine it with functional information to optimize recommendation performance. In this paper, we construct a mashup-tag-API graph based on tag similarity to address the sparsity between mashups and web APIs. We also design a method to jointly incorporate structural and functional information. Structural information is mined through multiple metapath aggregation mechanisms. Mashups are aggregated into different clusters based on structural information. Developers' requirement features are then extracted using functional semantic components and mapped to groups of similar mashups. Finally, the model recommends highly compatible network APIs based on historical call records from these groups. Comprehensive testing using real-world datasets from ProgrammableWeb demonstrates that our method outperforms baseline methods in terms of accuracy, recall, and MAP.