Web Service Recommendation based on Graph Attention Network (GAT-WSR)

In recent years, Web APIs have been greatly developed, which makes it convenient to develop Mashups from abundant sources of Web APIs. Nonetheless, it synchronously brings difficulties to choose suitable APIs for a Mashup. The existing web service recommender systems based on collaborative filtering only mine the historical behavior data of users, resulting in one-sided recommendation results. However, service recommendations should not only meet the functional needs of users, but also consider non-functional features to enhance the diversity of recommendations. In this work, we propose a method that can comprehensively use the information of functional semantics, non-functional features, and service invocation behavior. At the same time, we found that the relationship between Mashups and APIs is essentially a kind of graph data. By building a graph data network and directly inputting information into the GAT, nodes can extract their neighbors' information, so as to predict whether there are edges between nodes and complete the recommendation of APIs for Mashups. In view of this, this paper proposes a Mashup service recommendation model based on a GAT, which generalizes Mashups and APIs into network nodes, combines the text feature vector representation of nodes with other non-functional features. We use the GAT to capture the different weight contributions of different neighbor nodes, mix node information to output new node features. The extracted features are used for edge prediction tasks, so as to recommend appropriate APIs for Mashup applications. The results show that our model outperforms some state-of-art models in web service recommendations.