Research On Personalized Recommendation Algorithm Based On Tensor Decomposition

The recommendation algorithm is an important part of the network service platform.It can accurately and efficiently recommend favorite products according to the user's behavior and provide services that meet the user's needs.As we all know,Amazon's recommendation system provides users with an excellent user experience,where users can quickly buy their desired product.According to statistics,Amazon's 35% of sales is related to the recommendation system.With the rapid development of information technology,the number of users has gradually increased,and the network service platform has become the first choice for users to find services.Therefore,improving the efficiency and accuracy of the recommendati on algorithm is the primary task of the current recommendation system and it has always been a research hot issue for scholars.With the widespread adoption of Internet technology,users' interactions have become more frequent and the scale of network data has become increasingly large.Traditional recommendation algorithms has been unable to meet actual application requirements.There are two main reasons: first,some algorithms only consider the binary relationship between entities(users or products),and the links between multiple features are ignored.It is impossible to guarantee the accuracy of the recommended system in a large data environment.Second,without distinguishing the feature differences of the entities in the system,it is impossible to effectively identify the importance of different entities according to the relevant features,so that some potentially important content may be ignored.In this paper,we propose a user clustering personalized tag recommendation algorithm based on tensor decomposition for the traditional recommendation algorithms do not consider the entity differences in the system and ignore the multiple attributes of the data.In this algorithm,firstly,the user is clustered according to the difference of user's activity and similarity,so that the data set is divided into a number of small data sets.Secondly,the importance of each is calculated based on the relationship among users,products,tags and ratings,and the three-dimensional tensor model is constructed by combining the weights of four relationships.Finally,the Tucker decomposition method and the least-squares method are used to optimize the tensor model to generate a series of recommended results.The experimental results show that the proposed algorithm has a better improvement in recommendation accuracy than other recommended algorithms,which verifies the effectiveness of the proposed algorithm.