DINRec: Deep Interest Network Based API Recommendation Approach for Mashup Creation

摘要

Recommending appropriate APIs for Mashup creation has become a challenge as the number of APIs from different sources grows fast. In order to understand the relationships among multiple ecosystem APIs, most existing API recommendation methods focus on semantic similarity relationships but underutilize the composition and cooperation relationships between APIs, which may lead to low recommendation precision. In view of this problem, a Deep Interest Network based API Recommendation approach (DINRec) for Mashup development is proposed in this paper. In this approach, APIs are chosen incrementally for compositing into a Mashup and in that process the embedding vector of the Mashup's existing composition features will be updated adaptively by using Deep Interest Network. Moreover, a Doc2simu model is used to help training industrial deep networks with relatively small amounts of dataset. Finally, some experiments on real-world dataset are implemented to verify the efficiency of our proposed approach.