QF-RNN: QI-Matrix Factorization Based RNN for Time-Aware Service Recommendation

Driven by the widespread application of Service Oriented Architecture (SOA), the quantity of Web services and their users keep increasing in the service ecosystem. If historical service invocation records can be gathered and accumulated, it is meaningful to recommend suitable services that users may invoke in the near future. However, most existing recommend algorithms bear major limitation of not taking consideration of dynamic characteristics of both users and Quality of Service (QoS). To address this concern, this paper proposes a time-aware recommendation algorithm for runtime service selection. Firstly, a QoS Observation Matrix is created integrated with Invocation Record Matrix. Afterwards, matrix factorization is applied to extract user-preferences and service-features, respectively. Due to their dynamic characteristics, the Long Short Term Memory (LSTM) model is leveraged to learn and predict preferences and features. Finally, a service recommendation list is generated for users based on LSTM predictions. Experimental results on a real-world dataset show that the proposed algorithm outperforms baseline methods in terms of accuracy and recall.