Abstract
Predicting Web service interactions such as composition and substitution provides support for developers during mashup design. In this paper, we propose a deep-learning approach for predicting compositions and substitutions. To the best of our knowledge, this work is the first to adopt deep learning for interactions prediction. We use stacked autoencoders to learn latent service features. A deep feed forward neural network leverages the learned features and the history of previous interactions to predict new ones. We conducted extensive experiments on real-world Web services to illustrate the performance of our approach. We show that the use of deep learning achieves a high accuracy level and outperforms existing models such as multi-layer perceptron and support vector machine.