An effective automatic update approach for web service recommender systems based on feedforward-feedback control theory

With the rapid development of Web services, designing effective service recommendation technologies is becoming more and more important. Recently, Collaborative Filtering (CF) has become a mainstream approach for service recommendation, by predicting missing QoS (Quality Of Service) values for candidate Web services. However, CF algorithms are usually evaluated in a static context. In reality, a Web service recommender system inevitably experiences a continuous influx of new training data. CF will suffer a performance degradation if new training data are not timely considered for retraining. But too frequent retraining will bring a heavy computation overhead. In order to balance the system performance and the computational cost, we utilize a feedforward-feedback controller for automatic system updating. Experimental results demonstrate that this controller can effectively deal with both the performance deviation within the system and the primary observable disturbance from outside the system, thus to maintain a satisfactory system performance.