Research On Collaborative Filtering Recommendation Algorithm Based On Label Classification And Trust Auto-Encoder

With the rapid development of the Internet,people's ability to produce,copy and disseminate information has been greatly enhanced.The whole society is facing an unprecedented problem of information overload.Personalized recommendation system is a powerful method to solve this problem.In the field of recommendation systems,collaborative filtering is one of the most diffusely used means.CF method uses user item scoring matrix.In practical applications,recommendation accuracy is greatly reduced because the scoring matrix contains a large number of missing values,and for new users and There is a cold start for new items.In recent years,deep learning is diffusely applied in the fields of natural language processing,audio recognition and computer vision,and has received rapid development,which also brings new opportunities for recommendation systems.In this paper,the traditional collaborative filtering algorithm is improved firstly.The label information is used to help mitigate the problem of collaborative filtering sparseness.A collaborative filtering recommendation algorithm LCCF based on label classification is proposed.The incomplete data samples are classified according to the tags to make the matrix of decomposition.Depending on the class,the iterative projection pursuit method is used to calculate the linear combination of the dependent matrix and its corresponding weights,and finally the collaborative filtering recommendation.Aiming at the limitation of traditional collaborative filtering and the single information of scoring information,combined with the denosing autoencoder in deep learning model,a collaborative filtering recommendation algorithm TDAE based on trust information is proposed,which extracts hidden by correlation calculation.The trust information is then integrated with the explicit trust information and scoring information in the data set and the denosing autoencoder model,and the input of the denosing autoencoder is thinned,and finally the collaborative filtering recommendation is performed.