Research And Implementation Of Multi-scene QoS Prediction And Service Recommendation Model Based On Attention Mechanism

The popularity of cloud computing technology has provided developers of services with a transparent and reliable infrastructure,and has provided users with convenient and efficient services of all kinds.The development of mobile Internet has greatly enhanced users’ demand for services,making the number of Web services proliferate.Faced with the huge number of services,it is difficult for users to select quality services that meet their needs in time.As the number of services grows,the service functions gradually become saturated,and a large number of services with the same or similar functions appear,it is difficult to effectively distinguish the merits of services by considering only their functions,and QoS,which represents the nonfunctional attributes of services,becomes a key indicator for measuring services.However,users cannot invoke all the services to obtain the corresponding QoS and select the best service to access from them,so it is necessary to carry out research related to service recommendation models based on QoS prediction.In this paper,by using deep learning algorithms such as attention mechanism,we build a service recommendation model based on QoS prediction in multiple scenarios,which improves the accuracy of QoS prediction while meeting the needs of different service recommendation scenarios,and the main work of this paper is as follows:For the initial cold start scenario of users,an attention-based multi-order feature interaction learning service recommendation model is proposed.In feature processing,similar user and service features are obtained using historical invoke records and distance weights,while static contextual features such as location and network are introduced to alleviate cold start.For feature learning,on the one hand,a locationaware attentional factorization machine is constructed to learn linear and low-order feature interactions by using personalized hidden vector weights for different feature interactions through an attention mechanism.On the other hand,a higher-order feature learning algorithm based on the multi-headed attention mechanism is introduced to further learn higher-order combinations between features.Eventually,the two are integrated to achieve multi-order interaction learning for different features as well as combinations,and this is used as a basis to achieve unknown QoS prediction for relevant service recommendation.In the time-aware service recommendation scenario,a service recommendation model based on time-aware dynamic QoS prediction is constructed.Firstly,the data of time series are encoded by time-dependent Time2 Vec,which preserves the periodic characteristics of the series data itself.Secondly,the QoS data of the first k time slices are learned based on LSTM as well as the self-attention mechanism to capture the potential connections between them and obtain the final sequence representation.Finally,the personalized score weights generated by combining user and service information are then combined to generate QoS prediction values for the current time slice,while the accuracy of QoS prediction and service recommendation under timeaware scenarios is improved by continuously iterating the model by passing the data back.For the proposed multi-scenario QoS prediction and service recommendation model,extensive experiments are conducted based on public available datasets to verify that the proposed model in this paper outperforms other models in terms of prediction and recommendation accuracy,and can be applied to initial cold-start and time-aware scenarios by comparing classical and latest service recommendation models.The effectiveness of each component of the model is further verified through experiments on different parameters and components of the model,and the influence of relevant parameters on the effect of QoS prediction and service recommendation is analyzed.Finally,this paper constructs a prototype of a multi-scene QoS prediction and service recommendation system based on the attention mechanism,and verifies that the model can be effectively applied in the actual system to improve the performance of QoS prediction and service recommendation.