Research On Collaborative Filtering Recommendation Algorithm Based On Attribute Similarity Latent Factor Model Variable Weight Fusion

With the continuous development of information and Internet technology and its applications,not only has the amount of data exploded,but the data structure has become more and more complicated.Faced with the negative impact of "information overload" that makes it difficult to analyze,judge,select,and use the required information,people introduce information filtering technology based on personalized recommendations on the basis of search engines.Among many personalized recommendations,collaborative filtering recommendation algorithms are the most commonly used,and among them,latent factor model collaborative filtering recommendation algorithms are the most typical.However,in the process of decomposition and dimensionality reduction of the score matrix,some hidden feature information will be lost,which will affect the accuracy of recommendation.Therefore,based on the similarity transitivity,an latent factor model based on(user and item)attributes similarity is established to make up for the missing part of the hidden feature information.In addition,different collaborative filtering recommendation algorithms have different advantages and disadvantages.Based on the comparable fusion of data,a fusion model based on variance and minimum variable weight is established to balance the differences in the extraction of scoring matrix features by different collaborative filtering recommendation algorithms.Based on the attribute-based similarity-implicit semantic model and the variance-based and minimum variable-weight fusion model,a "cooperative filtering recommendation algorithm based on attribute-like latent factor model and its variable-weight fusion" is proposed.In this thesis,by identifying the current status of domestic and foreign research on collaborative filtering recommendation algorithms and the basis for the significance of the subject research,the target content and technical route of the subject research are constructed;then the basic knowledge of personalized recommendation is discussed,and collaborative filtering based on latent factor model is analyzed.The basic idea,technical structure and process method of recommendation algorithm and fusion recommendation algorithm.Based on this,thekey issues of the subject were researched,the basic idea and frame structure based on the attribute similarity implicit semantic model and the variance and minimum variable weight fusion model were proposed,and the data flow and collaborative filtering recommendation algorithm based on attribute similarity implicit semantic variable weight fusion fusion algorithm were constructed.Calculation process;and describe the algorithm in modules,and use Python3 to implement it.Through simulation experiments,the performance of the latent factor model.based on attribute similarity completely exceeds that of the latent factor model,which makes up for the defect of the hidden feature information of the latent factor model..The prediction correlation and prediction accuracy are improved.The hidden factor F is greater than or equal to 50 Later,the root mean square error gradually decreased and converged to the 0.94-0.95 interval.When F = 80,the latent factor model.based on attribute similarity is significantly improved compared with the traditional latent factor model.,the prediction accuracy is improved by 1.12%,and the prediction correlation is improved by 4.9%.Through variable weight fusion of multiple collaborative filtering algorithms,the attribute-based similarity-implicit semantic variable weighted fusion collaborative filtering recommendation algorithm has been improved on the basis of the attribute-based similarity-implicit semantic model,and the prediction accuracy is increased by about 0.9% when F = 80.The predicted relevance is increased by about 0.58%.