Research On Neighborhood-based Collaborative Filtering Recommendation Algorithm With Sparse Data

With the rapid development of the Web 3.0 social platform,the acquisition of network information has become very convenient and rapid.However,at the same time,the generation of a large amount of data information also seriously interferes with the user's acquisition of important information,and personalized recommendation technology is thus generated.The personalized recommendation technology helps to alleviate the negative impact of information overload.Collaborative filtering(CF)is A technique commonly found in personalized recommendation systems.Among them,neighborhood-based collaborative filtering is the most classic recommendation algorithm in CF algorithm.The core idea is to use a similarity measure to find the nearest neighbor set,and then use these sets,sum their scores by weighting,and then form a recommendation list.The accuracy of the similarity calculation determines the performance of the recommended algorithm.The existence of high-dimensional sparsity of data makes the traditional collaborative recommendation algorithm lack accuracy in the similarity measurement.In this paper,the high-dimensional sparsity problem of traditional algorithms is studied in depth from the aspects of scoring matrix filling,improved similarity and dimensionality reduction.This article mainly does the following work:(1)Research on collaborative filtering algorithm filled by scoring matrix.In this paper,the decomposition of the matrix is first performed by the Probability Matrix Decomposition method.Then,using the low-dimensional matrix obtained by the decomposition,the approximate scoring matrix is obtained inversely.The scoring matrix is filled only on the union of the evaluation items between users,and in order to preserve the features of the original sparse matrix,the similarity adjustment is performed by weighting,so that the similarity of the fusion not only considers the similarity between the users of the original sparse matrix,Moreover,the similarity between the users after filling is considered.At the same time,trust factor is added to constrain the sparse matrix filling,and the hypothesis of the filling algorithm is reduced.(2)Research on collaborative filtering recommendation algorithm based on fusion similarity.By improving and optimizing the similarity,the problem of high dimensional sparsity of data can be effectively alleviated.The fusion similarity can fully exploit the user's behavior characteristics,comprehensively consider the advantages of Jaccard similarity,Barthel coefficient(BC)and maximum information coefficient(MIC),and can make up for the shortcomings of the single similarity calculation method,which is a more comprehensive similarity model.The similarity after the fusion can fully consider the user's evaluation of the evaluated project,and can calculate the similarity without the common project score between users.(3)The collaborative filtering recommendation algorithm based on the combination of project heat and similarity is proposed.The traditional neighborhood-based collaborative filtering algorithm only considers the similarity factor and the negative value of traditional matrix decomposition.A new similarity measure method is proposed combining project heat and similarity.Firstly,the scoring matrix is filled,and the non-negative implicit feature space of the project is obtained by non-negative matrix factorization.Then,the similarity between the items is obtained in the non-negative implicit feature space of the project,and the first-stage neighbor set is generated.Then build the project heat model and integrate it into the similarity model,introduce the trust factors between projects into similarity space,find the nearest neighbor again,and complete the prediction recommendation.(4)Based on the proposed three new neighborhood-based collaborative filtering recommendation algorithms,the design of the movie recommendation system is completed.At the same time,in order to verify the functionality of the movie recommendation system,the movie recommendation system is tested to generate a recommendation list.