Attentional Matrix Factorization with Document-context awareness and Implicit API Relationship for Service Recommendation

The rapid development of the mashup approach significantly plays a pivot role in building Web-based and mobile applications. Easy accessibility of data and functions are the main advantages for programmers to develop mashups from abundant sources of APIs. However, it simultaneously brings more difficulties to choose suitable APIs for a mashup, especially when the historical relations between APIs and mashups are very sparse. Existing probabilistic matrix factorization (PMF) recommender systems can effectively exploit the latent features of the invocations with the same weight. However, not all features are equally significant and predictive, and the useless features may bring noises to the model. Also, many current works explored the influence of mashups’ relationships, but few of them sheds lights on the relationship between APIs and their contextual interaction, which can be mined from their content description. This paper improves the PMF model by distinguishing the importance of latent feature interactions. We present an Attentional PMF model, which leverages a neural attention network to learn the significance of feature interactions and uses Doc2Vec technique for mining the contextual information. We also exploit the relationship between APIs from both their contextual similarities and invocation history and add them to the prediction model as a regularization part. Our experiments are performed with datasets from ProgrammableWeb. The results show that our model significantly outperforms some state-of-art recommender systems in mashup service applications.