Research On Multi-level And Multi-feature Personalized Recommendation Model

With the advancement of modern science and technology,the total amount of data in all walks of life is constantly increasing,and it is difficult for many users to obtain effective information from the massive data.People have entered the era of information overload.Personalized recommendation has become one of the most effective means to solve information overload,and it is also a hot technology in the field of massive data mining research.However,in traditional recommendation algorithms,only the scoring information between users and items is often considered,and the basic information of users and items is missing,which makes it difficult to further improve the recommendation accuracy.In recent years,as deep learning technology shines in various fields,many scholars have applied deep learning technology in the field of personalized recommendation,and achieved certain results.However,few scholars have comprehensively considered the implicit interaction characteristics and latent characteristics of users and items.Therefore,this paper mainly studies the multi-level and multi-feature personalized recommendation model,and integrates deep learning technology to comprehensively consider the basic information of users and items,the implicit interaction features of users and items,and multi-level latent features,so that the performance of personalized recommendation can be further improved..The main research contents of this paper are as follows:1)The currently widely used personalized recommendation algorithms or models,as well as the related basic principles of deep learning technology,are introduced.It focuses on the recommendation algorithm based on collaborative filtering,the personalized recommendation model based on deep learning related to implicit interaction features and multi-level latent features,explains the main principles of the recommended algorithm or model and analyzes its advantages and disadvantages.2)A personalized recommendation model based on the interaction of explicit features and implicit features is proposed.The basic information of the user and the item is taken as the explicit feature,and the implicit interaction feature between the user and the item is obtained from the rating matrix of the user and the item.The initial interaction features are sent into a layer of fully connected neural network to enable deep interaction between explicit features and implicit features.Finally,a personalized recommendation model is established with the user's rating data on the item as a label,which realizes the relationship between the explicit features of users and items.Interactions between implicit features.3)A personalized recommendation model that considers multi-level latent features is proposed.Taking the basic information of users and items as the initial features,using FM,CN,WN and DNN,four feature mining algorithms to perform latent feature mining on the initial features,and then splicing the four levels of latent hierarchical features and sending them to a layer of full connection The interaction between multiple levels of latent features is strengthened in the neural network,and finally a personalized recommendation model is established with the user's rating data on the item as a label,which realizes the consideration of the multiple levels of latent features of the user and the item.4)A multi-level and multi-feature personalized recommendation model is proposed.The features obtained from the initial interaction between the explicit features and implicit features of the user and the item are used as the initial features,and the personalized recommendation model considering the multi-level latent features is used to mine the initial features with four levels of latent features,and use the user's rating on the item.The data is used as a label to establish a new personalized recommendation model,which realizes the consideration of multiple potential hierarchical features of the initial interactive features of explicit features and implicit features.Experiments on four public datasets show that the personalized recommendation model based on the interaction of explicit features and implicit features enables deep interaction between explicit features and implicit features.Compared with other models that consider implicit features,the model can The prediction accuracy of personalized recommendation is effectively improved.In addition,the influence of the dimension of the embedding layer and the number of neurons on the performance of the model is studied,and the time complexity of the model is briefly evaluated;the personalized recommendation model considering multi-level latent features is considered comprehensively4 Compared with the model considering single and two-level latent features,this model can effectively improve the prediction accuracy of personalized recommendation.In addition,the influence of the dimension of the embedding layer and the number of neurons on the performance of the model is studied,and the multi-level latent features are analyzed.The combination of hierarchical latent features is used for ablation experiments,and finally the time complexity of the model is briefly evaluated;however,the multi-level and multi-feature personalized recommendation model,due to the secondary latent feature mining after the initial interaction between features,uses the features learned from the initial interaction.The obtained feature relationship is masked,so that the model cannot mine the potential relationship between users and items well,so the model does not further improve the prediction accuracy of personalized recommendation.