Scholarlink

<p>Driven by service-oriented computing technology, the number of software services within the service ecosystem has grown rapidly. How to accurately and personally recommend suitable services for users from massive services has become a critical challenge. Existing service recommendation models suffer from data sparsity, making it challenging to thoroughly understand users’ preferences. Some methods attempt to mitigate this issue by integrating users’ social relationships and interaction histories. However, these methods ignore reviews, which makes it difficult to capture users’ fine-grained preferences, and fail to fully leverage the diverse relationships and their interactions to mine user preferences and service characteristics, which affects the comprehensiveness and personalization of the recommendation results. To address these issues, this paper proposes a Hierarchical Heterogeneous Graph Convolution Network and Improved LightGCN for Service Recommendation. This method effectively captures fine-grained user preferences and service characteristics contained in user reviews, while integrating multi-source heterogeneous information to improve recommendation performance. First, we construct a heterogeneous graph based on user-service interaction relationships, user social relationships, and service similarity relationships, incorporating user reviews on services as edge features to comprehensively reflect the complex semantics in the service ecosystem. Then, we introduce a gating unit into the LightGCN aggregation process to enable review-aware interaction modeling between users and services. Furthermore, a novel hierarchical heterogeneous graph convolutional network is designed to perform hierarchical modeling and cross-layer fusion of semantic information across various types of relationships. This approach enhances the accuracy and diversity of user and service representations, thereby improving recommendation performance. Extensive experiments on two real datasets have demonstrated HHGCN outperforms current mainstream recommendation techniques.</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

Hierarchical heterogeneous graph convolution network and improved LightGCN for service recommendation

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Zhiying Cao,
  • Miao Cheng,
  • Xiuguo Zhang,
  • Dezhen Zhang

摘要

Driven by service-oriented computing technology, the number of software services within the service ecosystem has grown rapidly. How to accurately and personally recommend suitable services for users from massive services has become a critical challenge. Existing service recommendation models suffer from data sparsity, making it challenging to thoroughly understand users’ preferences. Some methods attempt to mitigate this issue by integrating users’ social relationships and interaction histories. However, these methods ignore reviews, which makes it difficult to capture users’ fine-grained preferences, and fail to fully leverage the diverse relationships and their interactions to mine user preferences and service characteristics, which affects the comprehensiveness and personalization of the recommendation results. To address these issues, this paper proposes a Hierarchical Heterogeneous Graph Convolution Network and Improved LightGCN for Service Recommendation. This method effectively captures fine-grained user preferences and service characteristics contained in user reviews, while integrating multi-source heterogeneous information to improve recommendation performance. First, we construct a heterogeneous graph based on user-service interaction relationships, user social relationships, and service similarity relationships, incorporating user reviews on services as edge features to comprehensively reflect the complex semantics in the service ecosystem. Then, we introduce a gating unit into the LightGCN aggregation process to enable review-aware interaction modeling between users and services. Furthermore, a novel hierarchical heterogeneous graph convolutional network is designed to perform hierarchical modeling and cross-layer fusion of semantic information across various types of relationships. This approach enhances the accuracy and diversity of user and service representations, thereby improving recommendation performance. Extensive experiments on two real datasets have demonstrated HHGCN outperforms current mainstream recommendation techniques.