Web Service Recommendation Based on Time Series Forecasting and Collaborative Filtering

Quality of Service (QoS) has been widely used for personalized Web service recommendation. Since QoS information usually cannot be predetermined, how to make personalized QoS prediction precisely becomes a challenge of Web service recommendation. Time series forecasting and collaborative filtering are two mainstream technologies for QoS prediction. However, on one hand, existing time series forecasting approaches based on AutoRegressive Integrated Moving Average (ARIMA) models do not take the latest observation as a feedback to revise forecasts. Moreover, they only focus on predicting future QoS values for each individual Web service. Service users' personalized factors are not taken into account. On the other hand, collaborative filtering facilitates user-side personalized QoS evaluation, but rarely precisely models the temporal dynamics of QoS values. To address the limitations of existing QoS prediction methods, this paper proposes a novel personalized QoS prediction approach considering both the temporal dynamics of QoS attributes and the personalized factors of service users. Our approach seamlessly combines collaborative filtering with improved time series forecasting which uses Kalman filtering to compensate for shortcomings of ARIMA models. Finally, the experimental results show that the proposed approach can improve the accuracy of personalized QoS prediction significantly.