Personal Web API Recommendation Using Network-based Inference

In this paper, we evaluate a generic network-based inference algorithm for Web API recommendation. Based on experimental data collected from the Programmable Web repository, we construct two tripartite networks: one where the nodes are Web APIs, users and mashups, and another where the nodes are Web APIs, users and tags. Experimental results show that the network-based inference algorithm yields higher precision, ranking quality and personalization score when applied to the second network. This approach also outperforms three existing methods: a global ranking method, a collaborative filtering method and the Programmable Web recommendation tool.