Personalized Recommendation System Based On CTGAN-K Collaborative Filtering

Now is the era of high-speed Internet,the ways for users to create information and obtain information are more diverse and convenient.Various industries such as e-commerce economy,music recommendation,display and game recommendation are in full swing.At the same time,the generation speed of data has reached exponential growth which has jumped from TB level to ZB level.The purpose of this dissertation is to extract valuable information from these diverse but low-value data and find a more accurate personalized recommendation system for users.The CTGAN-K collaborative filtering algorithm is a hybrid collaborative filtering algorithm that combines GAN and cluster,which improve the recommendation accuracy and extensibility in this paper.The specific research contents are as follows.1.Based on the CTGAN-K model,the basic information of the user is combined with the basic properties of the behavior data and the behavior label.The current user portraits are roughly categorized according to the basic information that the user provides,such as gender and age.CTGAN can generate simulation data with Gaussian Mixture Model.It is possible to solve the problem of data cold-start and data sparsity in recommendation systems.2.Screening information kernel based on improved user portrait.The existing information kernel extraction of users is mainly heuristic strategy.Such results are generally locally optimal due to subjective influence.The hybrid information core extraction based on frequency and ranking proposed in this paper considers the user clicks to extract better core users.3.Combined with user item scoring algorithm,this paper presents the personalization of recommendation system by combining the scores of different documents in recommendation.Finally,the author evaluates and compares several recommended algorithms,and gives the final personalized recommendation.