.png){kind=icon}
Return
Adaptive hypergraph contrastive learning for cloud API recommendation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.133144.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Cloud APIs, as the core carriers of service delivery, data exchange, and capability reuse, have become indispensable in contemporary software development and operations. However, with the proliferation of cloud APIs, developers face substantial challenges in efficiently identifying and selecting appropriate APIs. Most existing studies primarily focus on propagating and aggregating information along interaction paths between mashups and cloud APIs, capturing local collaborative signals for recommendation. Nevertheless, these methods suffer from two limitations: 1) Deeper propagation of information tends to induce over-smoothing, making Mashup and cloud API representations indistinguishable. 2) The lack of exploration of potential global dependencies among Mashups and cloud APIs makes it difficult to alleviate data sparsity in local interaction modeling. To address these limitations, we propose an Adaptive Hypergraph Contrastive Learning framework (AHCL), which jointly models local collaborative relationships and global dependencies. Specifically, in addition to the local interaction graph, we adaptively construct Mashup and cloud API hypergraphs to capture global dependencies. Moreover, we design a multi-level contrastive learning mechanism with an augmentation strategy based on node degree centrality to enhance consistency between local and global structural relations. Additionally, an exclusivity constraint is introduced to reduce the aggregation of functionally redundant cloud APIs and encourage complementary API recommendations. Extensive experiments on two real-world datasets demonstrate that AHCL consistently outperforms state-of-the-art methods. In particular, compared with the strongest baseline, AHCL improves HR@10/NDCG@10 by 7.95%/6.10% on PWA and 4.89%/3.58% on HGA, while also showing strong robustness under sparse interaction settings. The implementations are available at: https://github.com/MengMeng3399/AHCL.
Keywords:
Recommender systems
Cloud API
Contrastive learning
Hypergraph
Journal
IF:
7.5
Papers:
3.0W
Citations:
10.2W
Organization
Cited Papers
No cited papers available