Research On Movie Hybrid Recommendation Based On Semi-Supervised Dimensionality Reductionon

With the rapid development of the Internet and the advent of the Big Data,the information in.the contemporary film industry is exploding.On the one hand,the choice of entertainment provided to people is dazzling,most traditional movie sites only recommend popular movies,this recommendation is unable to meet the personalized needs of users.Therefore,how to effectively analyze data and achieve effective recommendation for personalized entertainment needs of users has become a key aspect of creating commercial value in the contemporary entertainment industry.On the other hand,facing complex high-dimensional data,the recommendation system has to pay a huge storage and processing cost.So how to implement data reduction and cut recommendation cost while maintaining movie recommendation accuracy to meet the user's personalized entertainment needs has become a hot topic to be researched.In view of the two points,this paper studies and implements movie hybrid recommendation based on semi-supervised dimensionality reduction.The specific research contents are as follows:(1)In order to improve the recommendation accuracy to meet the user's personalized entertainment needs,this paper proposes a"Two-direction Classified Filtering Algorithm based on Negative Feedback and Time Penalty Term",which uses the negative feedback of popular movies mining potential information to fulfill available datasets.A Time Penalty Term is introduced at the same time,and the user preference weight is effectively adjusted according to the influence of time on the interest.At last,the user preference matrix and the movie feature matrix are clustered,the results will be filtered after two-direction clustering.The experiment proves that the "Two-direction Classified Filtering Algorithm based on Negative Feedback and Time Penalty Term" proposed in this paper can effectively improve the recommendation accuracy and meet user's personalized needs.(2)For the processing cost of high-dimensional data in the movie recommendation system,this paper proposes a "Locally Linear Embedding Algorithm based on Semi-supervised Repelling and Structure and Distance Measuring",it uses available tag information through semi-supervised methods to repel the unrelated data in neighborhood dataset.And faced with the limitation of the Euclidean distance,a comprehensive measurement method of Euclidean distance and Geodesic distance is proposed,which effectively improves the dimension reduction effect.It is proved by experiments that the "Locally Linear Embedding Algorithm based on Semi-supervised Repelling and Structure and Distance Measuring" can effectively cut the recommendation processing cost while maintaining the recommendation accuracy.(3)This paper has practiced the movie recommendation system,focusing on four important functions of the recommendation system:User identification function,User and movie information acquisition function,New users' interest determination function,Movie recommendation function,for the above four important functions,the movie recommendation system is successfully designed and implemented.Future follow-up studies will take the diversity of recommendation into account to attain higher user satisfaction and fully optimize the recommendation system.