Research On Personalized Recommendation Algorithm Based On User And Item Information

With the rapid development of artificial intelligence,mobile Internet and other technologies,the innovation of information technology brings great convenience to people's lives,but also often make users at a loss.In recent years,recommendation systems have been proven to be an effective way to solve the problem of information overload,and have been extensively studied.In recent years,hybrid recommendation algorithms have been widely studied because it can improve the shortcomings of a single algorithm by integrating multiple algorithms.It generally introduces auxiliary information of users and items to alleviate the data sparsity and cold start problems often faced in this field,and the key challenge of using auxiliary information is to effectively model the interaction relationship between features.A large number of studies have been carried out in recent years to use deep neural network(DNN)to learn the nonlinear interaction relations among features.However,the current research does not consider the different treatment of the interaction between different features,and the model structure is still imperfect.In addition,knowledge graph is also one of the sources of auxiliary information.The current related research has not paid attention to the low-order linear interaction between shallow features.In view of the above shortcomings,this thesis carried out in-depth research.The main work of this thesis are as follows:1.In this thesis,the existing FM-based deep network recommendation algorithm and the knowledge graph-based recommendation algorithm are deeply studied,and the advantages and disadvantages in the existing research are analyzed and summarized accordingly.2.Based on the idea of deep neural network learning features interaction,a new feature interaction network model(DPN)is proposed.The current model of using deep neural network to learn the interaction between features does not consider to differently treat the interaction between different features.In order to highlight the interaction relationship of important features,this thesis proposes a deep pruning network,which can better capture the high-order nonlinear relationship of the linear interaction between the important features.3.Based on the idea of FM combined with the feature interaction network(DPN),a new FM-based deep pruning network recommendation algorithm(DPFM)is proposed.Existing FM-based deep network recommendation algorithms cannot combine the serial and parallel structures of the FM module and the DNN module.In order to solve this problem,this thesis proposes the DPFM algorithm,which can learn the low-order linear relationship and important higher-order nonlinear relationship between features at the same time.4.Based on the idea of knowledge graph,a new fusion FM and knowledge graph recommendation algorithm(MKFM)is proposed.Most current knowledge graph-based recommendation algorithms do not focus on the low-order linear interaction between the shallow features of the user and the item.In view of the existing shortcomings,the algorithm proposed in this thesis uses the FM module to mine the shallow feature interaction between user and item features,which is used to enhance the representation ability of the recommendation model.5.In this thesis,a variety of advanced personalized recommendation algorithms are compared and analyzed on multiple public real data sets,and the effectiveness of the proposed algorithm is verified.