HSG-CDM: A Heterogeneous Service Graph Contextual Deep Model for Web Service Classification

Vivek Govind Lad, Eduardo Lima, Xumin Liu

Abstract: Service classification plays an important role in the field of Service-Oriented Computing (SOC). Significant research efforts have been made over the past years focusing on service classification, either adopting advanced NLP models to build service feature space from their textual attributes (such as service descriptions and titles), or leveraging graph models to learn service representation from structural information built from the relationships among services and other related entities such as users and mashups. This paper proposes a hybrid model, a Heterogeneous Service Graph Contextual Deep Model (HSG-CDM), that combines NLP models and graph deep models to capture both structural and textual features learned from service data. More specifically, we model service data as a heterogeneous information graph to capture the rich information of services, textual and structural, and adopt advanced deep learning techniques, such as Graph Convolution Network (GCN) and Bidirectional Encoder Representations from Transformers (BERT), to learn contextual service embeddings for classification. We conducted a comprehensive experimental study on a real world dataset to assess the performance of HSG-CDM. The results showed that HSG-CDM outperforms existing deep learning (DL) based models for service classification, such as ServeNet and Dual-Graph, as well as other non DL based approaches.