Research On Recommendation Method Based On Label Clustering And Multidimensional Trust Relations

With the rapid development of social networks,a huge amount of data information has been generated,and the research on related recommendation algorithms for tag data has emerged as a result,and has become one of the research hot issues.The impact of social networks in various institutional areas is growing,and it is imperceptibly changing and changing people's lifestyles,learning styles and even growth patterns.The emergence of a personalized recommendation system brings more user-friendly recommendations to users.Break through the traditional monotonous repetition of recommendations,and increase the diversity and novelty of recommendations.By mining the user's long and short interest images in multiple dimensions,it is possible to accurately integrate the pulse user needs and multi-strategy to optimize the diversity,novelty and timeliness of the recommendations.Researchers also have the opportunity to conduct analysis and research on online social networking platforms with large amounts of data,and have achieved high-quality,high-quality research results.For the traditional personalized recommendation system,only the static interest dimension is considered,and the problem of dynamic change of interest is not considered.This paper proposes a recommendation method of multi-faceted trust relationship based on tag clustering.This method draws on the idea of Ebbinghaus' s forgetting curve and traces it.The impact of changes in interest on user recommendations.Through the SPSS statistical analysis tool,the curve estimates a power function curve with a fairly high degree of fitness with the forgetting curve.In order to solve the problem of time accuracy and recommendation accuracy,a time-frequency window based time-frequency interest change function is proposed.Then,based on the redundancy problem of tag data redundancy,the spectral clustering algorithm is improved and the recommendation accuracy is improved.The tag cluster is used as the basis for dividing multiple facets to construct the untrusted tensor,and then the trust strength.Computational problems,by improving trust and non-trust relationships between clusters and clusters,and combining them to construct untrusted tensors.Finally,the tensor decomposition method is used to obtain the Top-N recommendation results.The experimental results are compared.The results show that the proposed algorithm can reproduce the semantic relationship while considering the redundancy of the tag data and consider the factors of the user's interest change with time,and the recommendation accuracy is significantly improved.