Research On Collaborative Filtering Recommendation Algorithm Based On Generalized Matrix Decomposition

The construction of the global information society and the existence of information overload problems make it more and more difficult for people to find what they want in the most satisfactory way when they need it.People tend to have too many choices to choose from,and they need to seek help from countless possibilities to explore and filter their preferences.The existence of a recommendation system can shorten the time consumption of people to get information,so that the right information meets the right person.Collaborative filtering is a very important part of the recommendation system,and the traditional solution has made it shine.The combination of deep learning and deep learning can help the recommendation system to obtain auxiliary information while learning the interaction between user items.This paper mainly does the following research work on the basis of collaborative filtering:(1)Summarize the classic heuristic and model-based collaborative filtering algorithms.Collaborative filtering algorithm is one of the most commonly used algorithms in recommendation algorithms,and it is also the focus of this paper.This paper summarizes the classic heuristic algorithm and model-based classic collaborative filtering algorithm for solving collaborative filtering problems.(2)Personalized recommendation algorithm based on generalized matrix decomposition model.The algorithm replaces the traditional Latent factor model(LFM)with a neural network,and uses the low-dimensional representation of the user and the item in the embedded layer to simulate the potential factor vector of the user and the item,which is directly used to learn the interaction between the user and the item.relationship.Based on the traditional generalized matrix decomposition model,this paper introduces the embedded layer to learn the potential feature representation of users and movies,and gives theoretical derivation based on explicit and implicit models.(3)Joint model based on generalized matrix decomposition model.Based on the improved generalized matrix decomposition model,the algorithm constructs a joint model based on the generalized matrix decomposition model from the perspective of the serial structure and parallel structure of the deep neural network,and introduces the multi-layer perceptron model to learn the potential features between user items.The nonlinear interaction part,the linear interaction component between the user and the potential factor feature between the items represented by the generalized matrix decomposition model is combined with the nonlinear interaction part between the user and the film potential factor of the multi-layer perceptron model learning,and the composition is based on A joint model of generalized matrix decomposition models.In this paper,the theoretical derivation based on the serial structure and the joint model based on parallel structure,the construction and training of the deep neural network model are given respectively,and the feasibility and effectiveness of introducing nonlinear interactions for the potential features of users and movies are verified.Moreover,based on the joint model based on serial structure and the joint model based on parallel structure,this paper introduces the combination of similarity matrix and joint model to realize Top-N recommendation.