A carregar...
Publicação
NCOHA-WSC
Autores
Orientador(es)
Resumo(s)
With the rapid growth in the number of Web services, accurate and efficient Web service classification has become crucial for improving the quality-of-service discovery. However, existing classification approaches often overlook the issues of noise and class overlap inherent in Web service data, which leads to degraded classification precision. To address these challenges, this paper proposes NCOHA-WSC, an approach for Web service classification designed to handle both noise and class overlap and to be easily integrated with existing machine learning–based service classification models. Specifically, noisy samples in the training data are filtered using confidence learning and information entropy, thereby reducing the negative impact of noise on the classification model during preprocessing. In addition, during the testing phase, the prediction results for overlapping services are corrected based on the label prior distribution, further improving classification precision. Experiments conducted on the real-world ProgrammableWeb dataset demonstrate that NCOHA-WSC is compatible with mainstream Web service classification models and can enhance the Macro-F1 performance of models such as ServeNet and CARL-Net to varying degrees. These results indicate that the proposed approach effectively mitigates the impact of noisy data on Web service classification and improves the precision of existing models in the presence of overlapping service classes.
Descrição
Zhang, F., Xue, L., Li, H., & Liu, C. (2026). NCOHA-WSC: Handling Noise and Class Overlap in Web Service Classification. Journal of Computational and Cognitive Engineering. https://doi.org/10.47852/bonviewJCCE62028810
Palavras-chave
Web service classification noise handling class overlap confidence learning prior distribution