Open APIs recommendation with an ensemble-based multi-feature model

Recommending the appropriate APIs from a large volume of Open APIs to application developers both accurately and efficiently has become a challenging problem. Established work usually takes only one feature of Open APIs into account, which decreases the accuracy of recommendations. In order to overcome this problem, we propose an ensemble-based approach to Open APIs recommendation with a multiple feature model, which integrates both machine learning and deep learning and synthesizes a set of multiple features to accurately make the recommendation. This approach employs One-hot, Word2vec similarity and Matrix Factorization techniques to obtain the multiple features information respectively, then concatenates the features to obtain a Multivariate Information Feature (MIF) matrix, and leverages an optimized Gradient Boosting Decision Tree (GBDT) and Gated Recurrent Unit (GRU) for feature selection. GBDT is good at processing dense numerical features, while GRU is good at processing sparse categorical features, Finally, the results are synthesized to obtain a recommendation result. We have compared our approach with other four Open APIs recommendation approaches on the Programmable Web, and verified the effectiveness of our approach in precision, recall, F1-measure.