Jointly learning invocations and descriptions for context-aware mashup tagging with graph attention network

With the growing prosperity of the Web service economy, more and more mashups have been developed that combine multiple Web APIs to achieve more powerful functionalities to accommodate complex business requirements. Consequently, mashup tagging has become an emerging task that is essential for managing and retrieving enormous service resources. Most of the existing mashup tagging methods are limited in several critical aspects such as the lack of explicit modeling for high-order connectivity, the neglect of discriminating the different importance of neighbors related to mashups adaptively, and achieving less desirable performance. To address the above limitations, in this paper, we propose a Context-Aware method to learn invocations patterns and descriptions for Mashup Tagging, named CAMT. Specifically, we explicitly model the high-order connectivity with two-graph evolution patterns (including the mashup-API-tag graph and the mashup-API-word graph) based on a graph neural network, and recursively propagating embeddings from neighbors of the target node to update its representation. Finally, a multi-head attention mechanism is exploited to discriminate the importance of neighbors adaptively. Comprehensive experiments on the real-world dataset demonstrate the effectiveness of CAMT when compared with many state-of-the-art baselines. For example, we achieve 10.7/9.6/12.7/9.0% gains in terms of P@5 / R@5 / MRR@5 / NDCG@5 metrics for mashup tagging, respectively. In addition, our model can achieve not only higher accuracy but also higher diversity and lower computational overhead.