Graph4Web: A relation-aware graph attention network for web service classification

Software reuse is a popular way to utilize existing software components to ensure the quality of newly developed software in service-oriented architecture. However, how to find a suitable web service from existing repositories to meet requirements is still an open issue. Among others, web service classification is one of the most essential and effective means for web service recommendation. Previous studies have concerned this problem, but a critical issue, i.e., the semantic and syntactic information for the web service, is often ignored. To address such an issue, in this work, we propose Graph4Web, which uses a relation-aware graph attention network for web service classification. Specifically, we first parse the web service description sequence into the dependency graph and initialize the embedding vector of each node in the graph by tuning the pre-trained BERT model. We further propose a relation-aware graph attention layer to learn and update the node embedding vector by aggregating the information of neighborhood nodes and the distinct types of relationships between nodes. In addition, we introduce the self-attention mechanism to acquire the high-level global representation for web service classification. Various experiments demonstrate that Graph4Web has better classification performance compared with seven baseline methods with three indicators.