Concept drift-aware temporal cloud service APIs recommendation for building composite cloud systems

摘要

The booming advances of cloud computing promote rapid growth of the number of cloud service Application Program Interfaces (APIs) published at the large-scale software cloud markets. Cloud service APIs recommendation remains a challenging issue for a composite cloud system construction, due to massively available candidate component cloud services with similar (or identical) functionalities in the cloud markets. As for a specific user, the probability distribution of the data indicating his/her preferences to the cloud service APIs may change with time, resulting in concept drifting preferences. To adapt users' preference drifts and provide effective recommendation results to composite cloud system developers, we propose a concept drift-aware temporal cloud service APIs recommendation approach for composite cloud systems (or CD-APIR) in this paper. First, we track users temporal preferences through users' behavior-aware information analysis. Second, we utilize Singular Value Decomposition (SVD) method to predict the missing values in the user-service matrices. Third, we identify the degree of users preference drifts by Jensen-Shannon (or JS) divergence. Finally, we recommend cloud service APIs by presenting a piecewise trading-off equation. Experimental evaluations conducted on WS-Dream dataset demonstrate that the CD-APIR approach can effectively improve the accuracy of cloud service APIs recommendation comparing with 7 representative approaches.