Scholarlink

<p>Web service recommendation plays a crucial role in modern service-oriented platforms. The challenge of accurately recommending Web services that meet developers’ needs, particularly in environments characterized by data sparsity and noise interference, has remains a critical challenge. Recent studies have demonstrated the effectiveness of self-supervised learning (SSL) in mitigating the impact of data sparsity and noise. Building on this insight, researchers have begun exploring the application of SSL in Web service recommendation systems, leveraging self-supervised signals to enhance model generalization and improve recommendation accuracy. However, not all multi-category associations established between Mashups and APIs exhibit substantive implications for service interactions, where irrelevant categorical relationships may introduce extraneous noise into these interactions. Therefore, a Web service recommendation method based on denoising self-augmented learning is proposed in this paper. The method not only preserves useful service interactions but also enhances the recommendation accuracy of Mashup-API pairs through an adaptive semantic alignment mechanism, enabling cross-view knowledge transfer within the embedding space. Specifically, the model constructs a three-view graph neural network that encodes the Mashup collaboration graph, API association graph, and service interaction graph, capturing their complex interrelationships. To further mitigate the negative impact of noise on the service recommendation system, this paper also introduces a denoising self-augmentation module, which effectively filters out service interaction noise through the cross-view alignment mechanism. Finally, the proposed model combines positive and negative sample comparisons to further improve the representation learning between services. The experimental results on real datasets show that compared with baseline methods such as GraphSAGE, LightGCN, HGT, Heco, SMIN and HGCL, the proposed model achieves improvements of 33.5%, 46.9%, 23.82%, 8.3%, 16.2%, and 7.26% in terms of HR, successfully validating the effectiveness of the approach.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DSRec: Denoising self-augmented learning for web service recommendation

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Buqing Cao,
  • Yating Yi,
  • Xiang Xie,
  • Qian Peng,
  • Lei Jiang,
  • Yijiang Zhao

摘要

Web service recommendation plays a crucial role in modern service-oriented platforms. The challenge of accurately recommending Web services that meet developers’ needs, particularly in environments characterized by data sparsity and noise interference, has remains a critical challenge. Recent studies have demonstrated the effectiveness of self-supervised learning (SSL) in mitigating the impact of data sparsity and noise. Building on this insight, researchers have begun exploring the application of SSL in Web service recommendation systems, leveraging self-supervised signals to enhance model generalization and improve recommendation accuracy. However, not all multi-category associations established between Mashups and APIs exhibit substantive implications for service interactions, where irrelevant categorical relationships may introduce extraneous noise into these interactions. Therefore, a Web service recommendation method based on denoising self-augmented learning is proposed in this paper. The method not only preserves useful service interactions but also enhances the recommendation accuracy of Mashup-API pairs through an adaptive semantic alignment mechanism, enabling cross-view knowledge transfer within the embedding space. Specifically, the model constructs a three-view graph neural network that encodes the Mashup collaboration graph, API association graph, and service interaction graph, capturing their complex interrelationships. To further mitigate the negative impact of noise on the service recommendation system, this paper also introduces a denoising self-augmentation module, which effectively filters out service interaction noise through the cross-view alignment mechanism. Finally, the proposed model combines positive and negative sample comparisons to further improve the representation learning between services. The experimental results on real datasets show that compared with baseline methods such as GraphSAGE, LightGCN, HGT, Heco, SMIN and HGCL, the proposed model achieves improvements of 33.5%, 46.9%, 23.82%, 8.3%, 16.2%, and 7.26% in terms of HR, successfully validating the effectiveness of the approach.