Web Service Recommendation Via Combining Topic-aware Heterogeneous Graph Representation and Interactive Semantic Enhancement
Buqing Cao,Qian Peng,Xiang Xie,Zhenlian Peng,Jianxun Liu,Zibin Zheng
DOI: https://doi.org/10.1109/tsc.2024.3418328
IF: 11.019
2024-01-01
IEEE Transactions on Services Computing
Abstract:With the continually increasing number of Web services, it becomes a challenging task to efficiently and accurately provide Web services that meet developers' functional requirements. Existing heterogeneous graph-based service recommendation methods simply utilize the heterogeneous structural features of the service network and suffer from the missing and blurring of service interaction semantic information due to the characteristics of meta-paths. In fact, service node description documents contain fine-grained semantics generated by multifaceted topic-aware factors, but few efforts are committed to mining them. Therefore, a Web service recommendation method via combining topic-aware heterogeneous graph representation and interactive semantic enhancement is proposed in this paper. It employs an alternating two-step aggregation mechanism, including meta-path instance intra-decomposition and meta-path inter-integration, which uniquely aggregates topic-aware factors according to the inferred topic distributions while preserving structural semantics. Additionally, it introduces the topic prior knowledge guidance module to improve the quality of the inference's topic factors. Simultaneously, the method designs the interactive semantic enhancement module to address the missing and blurring of service interaction semantic information caused by meta-paths. The module explores complex interaction patterns among services and utilizes personalized knowledge meta-network to enhance contrastive learning of service interaction semantics, allowing the personalized knowledge transformer with adaptive contrastive enhancement. The experimental results on the real dataset of ProgrammableWeb show that compared with the other nine methods, the proposed method has better service recommendation performance on evaluation metrics HR and NDCG representing accuracy and satisfaction, respectively.