Research On Synthetical Collaborative Filtering Model Based On Dynamic Time Sequence

Collaborative filtering algorithms have been widely used in the field of recommendation systems today and deal with explosive information overload problems.However,with the complications of data problems,such as the potential for hidden features of data that are difficult to capture,the increasingly sparseness of data sets,and changes in the timeliness of data,the accuracy of traditional collaborative filtering algorithm recommendations has reached a relative bottleneck.The matrix factorization model and DNNs(Deep Neural Network)model,which incorporates dynamic timing characteristics,have important research value for solving the problems of simulated timeliness and implicit data feedback in the field of recommendation systems.The thesis takes a movie scoring data set with dynamic timing properties as the research object,and gives a recommendation result that accords with its preferences to a given target user.In order to facilitate the use of the recommended model in the analysis phase,the data set is first preprocessed.PCA(Principal Component Analysis)is used to select a part of the core data for pre-analysis,and data normalization and noise reduction processing is performed on the data set according to the analysis result.Then,the clustering method is applied to the data set after the preliminary processing based on the clustering method based on the density peak.After the special sample is isolated,similar user clusters and movie clusters are divided.According to the bottleneck problem faced by traditional collaborative filtering algorithms,this paper proposes and designs a matrix factorization decomposition model based on dynamic time series.It consists of a static information capture model and a dynamic model with dynamic timing factors.It simulates user preferences over time.The trend of the continuous change of the trend and the degree of popularity of the movie over time.Because the matrix decomposition method of the model has certain limitations,there is still a bottleneck in solving the complex interaction between the user and the movie.In this paper,through the comparative analysis of research and experiments,the hidden interactions in the data are simulated according to the ideas of deep neural networks to improve the deficiencies.This thesis proposes and designs a general framework for deep collaborative filtering based on the composite matrix factorization model,and proposes a MLP(Multi-Layer Perceptron)module to simulate the difficult-to-capture implicit interactions between users and movie projects,and fusion matrix decomposition.Comprehensively obtain the recommended results,using RMSE(Root Mean Square Error)to comprehensively analyze and evaluate the recommended results of the proposed method on the experimental data set.Through the horizontal comparison with the recommendation results obtained by the mainstream recommendation method on the experimental data,the results show that the proposed dynamic temporal-based compound collaborative filtering model has a certain effect on the accuracy of the recommendation results.