Scholarlink

<p>The rapid growth of Web services poses a significant challenge in selecting high-quality services for Mashup creation from a wide range of options. Existing service recommendation methods fall short in using text to describe contextual relationships and neighbor information. Therefore, this paper proposes a pre-trained representation and negative sampling-based service recommendation, denoted as REST. Firstly, a multi-granularity semantic integration mechanism is designed that combines sentence-level textual representations with a global attention mechanism, enabling fine-grained semantic modeling of service descriptions and effectively capturing the key features of service functions. Next, the collaboration between two-way sequence modeling and the attention mechanism is utilized for context-aware service classification to obtain high-quality service representations. Finally, we investigate higher-order neighbor relationships between services and integrate interaction information into non-interacting services. By aggregating neighbor information from different layers of non-interacting services, we generate hard negative samples for model training. This approach effectively enhances the model’s discriminative capability between positive and negative samples, ultimately improving recommendation performance. We conduct extensive experiments on real service datasets from the ProgrammableWeb platform and show that REST outperforms the state-of-the-art methods in Recall, NDCG, and Hit.</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

Pre-trained representation and negative sampling-based service recommendation

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Ziming Xie,
  • Buqing Cao,
  • Yanxinwen Li,
  • Shangpeng Liu,
  • Guosheng Kang,
  • Zhenlian Peng

摘要

The rapid growth of Web services poses a significant challenge in selecting high-quality services for Mashup creation from a wide range of options. Existing service recommendation methods fall short in using text to describe contextual relationships and neighbor information. Therefore, this paper proposes a pre-trained representation and negative sampling-based service recommendation, denoted as REST. Firstly, a multi-granularity semantic integration mechanism is designed that combines sentence-level textual representations with a global attention mechanism, enabling fine-grained semantic modeling of service descriptions and effectively capturing the key features of service functions. Next, the collaboration between two-way sequence modeling and the attention mechanism is utilized for context-aware service classification to obtain high-quality service representations. Finally, we investigate higher-order neighbor relationships between services and integrate interaction information into non-interacting services. By aggregating neighbor information from different layers of non-interacting services, we generate hard negative samples for model training. This approach effectively enhances the model’s discriminative capability between positive and negative samples, ultimately improving recommendation performance. We conduct extensive experiments on real service datasets from the ProgrammableWeb platform and show that REST outperforms the state-of-the-art methods in Recall, NDCG, and Hit.