Multi-Label Web Service Classification Using Neural Networks

With the development of the Internet, web services have played an increasingly important role in the software industry in recent years, and the increasing number of web services has brought about problems such as increased difficulty in obtaining labels. Therefore, finding a technique that can automatically match and recommend the most accurate labels quickly and accurately based on the service descriptions published by service providers has become a hot issue in the field of service computing. The current web service classification methods mainly focus on the main labels of web services, and there is no good method to classify web services with multiple labels. In this paper for the problem of multi-label classification of web services, we propose a multi-label classification model BCLAS for web services based on label embedding and attention. The BCLAS model uses BERT to generate word vector sequences of service descriptions and labels, and uses stacking networks of CNN and Bi-LSTM to extract semantic features in service descriptions, after which the obtained semantic features and labels are jointly learned in the embedding space, the learned attention is weighted to the final web service description text sequence and label representation, the feature vector representation of the optimized text, the labels are also learned to different text feature representations, and finally the classifier classifies web services using the weighted label embedding representation.