Research Of Service Recommendation In Cyber-Physical Systems Based On Time And Location

With the continuous improvement of cyber-physical systems,the amount of services increases extremely.However,how to select appropriate services recommended to users has become an essential topic.To deal with different application scenarios,two models,LBFFM and LSTM-HPLT,are structured in this paper.Firstly,a location-aware service recommendation model based on field-aware factorization machines(LBFFM)was proposed.The model provides a distance weight to reduce the influence of location information when calculating the similarity between two users or services.Aiming at the difficulty of recommending services to the new user and the difficulty of recommending the new service to users,our model transforms the data into one-hot vectors and proposes a similarity calculation method based on distance weight.Furthermore,the field-aware factorization machines improve the prediction accuracy through learning the connection between the character and other fields.After that,a hybrid service recommendation model based on long short-term memory(LSTM-HPLT)was provided for the accuracy and performance of service recommendation in the sparse environment.The location and time information are adopted which can alleviate the data sparsity through filling in the missing or future QoS values in the adjacent time period.Our model selects similar users or services through the personalized similarity measure and their respective positions and then predicts the QoS value through the LSTM model trained by the QoS values created by similar users or services.Finally,extensive experiments have been carried out based on real QoS data in cyber-physical systems.According to the experiment results,the factor was verified that the two models proposed in this paper both have better performances than other service recommendation methods in NMAE or RMSE.Furthermore,a service recommendation system based on LBFFM and LSTM-HPLT was built to manage the information of users and services.The system can also provide personalized service recommendation in different scenarios.