Spatio-temporal context-aware collaborative QoS prediction

Abstract With the exponential growth of Web services, various collaborative QoS prediction methods have been suggested to make an efficient evaluation of quality-of-services (QoS) and assist users selecting appropriate services. It is still a technical challenge to be taken into account the impact of complex spatio-temporal contexts of service invocations and make use of their characteristics in the forecasting process. To this end, we propose two universal spatio-temporal context-aware collaborative neural models (STCA-1 and STCA-2) to make QoS prediction by considering invocation time and multiple spatial features both of service-side and user-side. Our proposed models utilize hierarchical neural networks to realize the embedding expression of original features, the generation of second-order features, the fusion of first-order and second-order features, the interaction between spatial features and temporal features layer by layer. In particular, attention mechanism is introduced to automatically assign weights to spatial features and realize the discriminative application in feature fusion. Experiments on a large-scale dataset demonstrate the effectiveness of the proposed method: (1) The prediction error can be significantly reduced in comparison with the baseline methods particularly in the case of sparse training data, where our models achieve a performance improvement by about 10.9–21.0% in term of MAE and NMAE, and by 2.4–7.8% in term of RMSE. (2) Attention mechanisms enable us to give intuitive explanations of the effectiveness of feature fusion more reasonably and thus strengthen the interpretability of the prediction models.