Research On Data Prediction Based On Filtering Fusion

In this paper,the data prediction method based on filtering fusion is the fusion of collaborative filtering algorithm and content filtering algorithm.In this paper,the data prediction based on filtering and fusion uses user historical behavior information,item attribute information and user's own characteristic information to predict users' potential unknown behavior tendency.The prediction method studied in this paper solves the matrix sparsity problem and cold start problem of the collaborative filtering algorithm to a certain extent,and also solves the problem of insufficient interactive information between users and items of the content filtering algorithm,so that the prediction accuracy is correspondingly improved and In some cases,the complexity of the model will be reduced.The main work of this paper is as follows:(1)Investigated several mainstream technologies of the current recommendation system: K nearest neighbour,collaborative filtering and content filtering,and compared their performance and complexity based on the same data set;(2)On the basis of the above work,the problems of the single algorithm are improved,the concept of "fusion coefficient" is proposed,and the filtering and fusion research based on single-type features is proposed to simplify the network and improve the accuracy of prediction.Insufficiency of the algorithm-improve the matrix sparsity problem,cold start problem and interactive information problems,and analyze the experimental results based on the public data set to give corresponding conclusions;(3)According to the aforementioned filtering and fusion algorithm based on singletype features,we propose another derivative filtering and fusion algorithm based on full features,which can provide more comprehensive user and item profile information,which can further improve forecast accuracy.At the same time,we also conduct experimental verification based on the public data set,and analyze the experimental data to give corresponding conclusionsThe purpose of this article is to integrate content filtering algorithms and collaborative filtering algorithms through the definition of sparse coefficients,and to predict unknown ratings by constructing a higher-order graph filter model.At the same time,we did not set the number of neighbors to be taken as a fixed value,so it can be adjusted adaptively according to the similarity between the user and the product.Experimental results based on public data sets(see Chapters 2,3,and 4 for details)show that the single-type feature-based filtering and fusion algorithm and the full feature-based filtering and fusion algorithm proposed in this paper can reduce the complexity of the recommendation system to a certain extent It also improves the prediction accuracy,and also provides a solution to the cold start problem of a single collaborative filtering algorithm.