Research And Implementation Of Personalized Recommendation System Based On Situation Awareness

At the same time as people obtain information content,it is easy to diverge from their own target content to more massive data,and thus cannot obtain the information they need in a timely manner.This kind of information overload problem has brought us a great burden of information.Not only has it not increased productivity due to Internet technology,but it has also reduced the productivity due to the huge amount of information.At this time,a message push service model has emerged.The recommendation system is the most important.One,with personalized services as the main appeal,taps the relationship between the user and the project and pushes information that may be needed but is difficult to actively acquire.With the further research of the researchers,it is found that the initial data size of the traditional recommendation algorithm widely used in e-commerce has a problem of cold start and sparsity,which leads to the problem of decreased accuracy.After researching the model prediction method based on content filtering,this paper puts forward a recommended method for constructing Bayesian network model based on context data and machine learning.The implicit variables are used to represent the user's preferences.Bayesian network construction for project scoring uses a hidden variable insertion algorithm based on the semi-cluster structure.The BIC score is used to select the optimal model.The EM algorithm calculates the conditional probability parameters of the project scoring Bayesian network model.Thus,context-sensing data is added to enrich the MovieLens data set,and then the RBNL model is used to predict user rating data.The hybrid recommendation strategy can synthesize the advantages of the two recommended components.In this paper,the RBNL model prediction algorithm and the traditional collaborative filtering algorithm are integrated.Using the idea of error prediction to infer accuracy,the error difference matrix is used to obtain the recommendation accuracy for each user and each item for each recommendation component.The adaptive weights are normalized to adjust the different recommended components in the mixed recommendation.In the impact,the error prediction matrix is processed in a certain amount,and then combined with the predicted result of the corresponding recommended component to be weighted to finally obtain a mixed recommended prediction score.Experiments show that the proposed method for predicting data from contextawareness data improves the accuracy of single-recommendation method without contextual data.However,the accuracy of collaborative filtering algorithm increases significantly after the amount of data in the later period of the system increases.The hybrid recommendation strategy of adaptive weights can synthesize the advantages of the two recommended components.At the initial stage of system operation,the model recommendation algorithm error is smaller,so the overall system prediction accuracy rate is better,and the accuracy of the collaborative filtering algorithm is higher in the middle and later stages of the system operation.Increased,and the system consumes less.Therefore,the adaptive weighted hybrid recommendation algorithm can supplement the advantages,overcome the defects caused by the data amount in different periods,and better improve the performance of the recommendation system.