A Web Services Classification Method Based on GCN

Abstract

With the development of Web2.0 and Web services, the diversity and amount of Web services are increasing quickly. Finding Web services to meet the needs of users has turned into increasingly difficult. It is a valid way to promote service discovery that classifying Web services with similar functionality. The existing Web services classification technology mainly focuses on exploiting the functional information, such as description texts or tags to achieve Web services classification, but ignores the network structure information implied between the words inside the Web service description text and the Web service description text itself. Therefore, we put forward an approach of Web service classification on top of graph convolutional neural networks. This method, firstly, uses the name, description text, tags of Web service as the basic corpus to construct a heterogeneous graph network of Words & Web service description according to word co-occurrence and word relationship among the Web service description document. In the heterogeneous graph network, all weights of the edges between the document nodes and the word nodes are calculated by used the term frequency-inverse document frequency, and the point mutual information is utilized to calculate the weights of the edges between different word nodes. Then, it applies the graph convolutional neural network to learn the embedding information of words and Web service description document, and transforms Web service documents classification problem into node classification problem. Finally, the experiment on ProgrammableWeb dataset display that the precision, recall, F-measure, purity and entropy of the proposed approach are greatly improved, compared to those of TF-IDF+LR, LDA, WE-LDA, LSTM, Wide&Deep, Bi-LSTM, Wide&Bi-LSTM.