Multi-information fusion based few-shot Web service classification

Automatic Web service classification becomes an essential topic in the services computing field. The distribution of Web services over various categories usually follows the long-tail distribution, suggesting that many categories (i.e., the tail categories) contain very limited services. An empirical experiment shows that the classification performance of tail categories is much worse than that of head categories due to the limited training samples. Existing works on Web service classification usually ignore this problem. Towards this issue, we propose a few-shot Web service classification approach called MIF-FWSC (multi-information fusion based few-shot Web service classification), which exploits both the knowledge learned from head categories and the information contained by category names to improve the classification of tail categories. Experiments show that our proposed approach for few-shot Web service classification achieves state-of-the-art accuracy on two real-world Web service datasets. (C) 2021 Elsevier B.V. All rights reserved.