Service Clustering Method Based on Knowledge Graph Representation Learning

With the changing of users’ requirements, the number of Web services is growing rapidly. It has been a popular research field to discover the suitable service accurately and quickly in service computing research. At present, most of the Web services published on the Internet are described in natural language. This trend is becoming more and more obvious. Existing service clustering methods are not only limited to a specifically structured document but also rarely consider the relationship between services into semantic information. In response to the problems mentioned above, this paper suggests a Service Clustering method based on Knowledge Graph Representation Learning (SCKGRL). This method firstly crawled the services data from ProgrammableWeb.com, use natural language tools to process the web services description document, and obtain the service function information set. Secondly, we constructed the service knowledge graph by using the service-related information, the triples are converted into vectors and minimize the dimension of service feature vectors due to the knowledge representation learning method. Finally, the services were clustered by the Louvain algorithm. The experiments show that SCKGRL gives better performance compared with other methods, such as LDA, VSM, WordNet, and Edit Distance, which provides on-demand service more accurately.