Research And Application Of Collaborative Filtering And Matrix Decomposition In Recommendation System

With the development of the Internet,people have more ways to obtain information,and the amount of information they obtained is also growing rapidly.While satisfying people's demand,mass information also brings troubles to people.A large amount of invalid information not only interferes with people's judgment on normal information,but also reduces the efficiency of information processing.The recommendation system is a better solution to deal with the massive information,which can process the information in a certain way and recommend the results that the user is really interested in,so that the user can improve the efficiency of obtaining the required information.The collaborative filtering algorithm generally estimates the user's preferences based on the user's evaluation information.However,due to the data sparse problem,in many cases,it is impossible to obtain the ideal recommendation results.In addition,the collaborative recommendation algorithm ignores the dynamic change of user interest.To address these issues,this paper proposes an improved collaborative filtering recommendation algorithm.The algorithm proposed in this paper mainly combines similarity transfer,user interest migration,matrix decomposition technology,etc.to solve the above problems.Firstly,this paper proposes a collaborative recommendation algorithm based on project similarity transfer.The algorithm improves the similarity calculation method.Firstly,the trust relationship is modeled on the project.Based on this,the similarity is transmitted,and then the similarity relationship between the two parts is obtained.Weighting to get a new project similarity and applying it to the project's score.Secondly,this paper proposed a latent factor model recommendation algorithm based on user interest migration.The algorithm introduces the time function,reconstructs the user's interest model,and implements the modification of the traditional model,and then uses the gradient descent method to solve.Finally,in view of the shortcomings of the above two algorithms,we proposed two kinds of fusion schemes to integrate the two algorithms: 1)the latent factor model in the collaborative algorithm is used,and the task of similarity calculation by forming the invisible features;2)The item-based collaborative filtering algorithm is integrated with the latent factor model algorithm by linear fusion.The results of the experiments show that the linear fusion method is better because it can supplement the missing information,adapt to the change of user interest,and greatly weaken a series of negative effects caused by sparse data.We applied the improved recommendation algorithm to the movie recommendation system.The movie recommendation system consists of three subsystems: the movie recommendation subsystem,the background management subsystem,and the user subsystem.The movie recommendation subsystem first collects and analyzes the user data,builds the user's hobby model,and then use the hybrid recommendation method to recommend the user's favorite movie to the other people.