Research On Implicit Video Data Attention Detection And Recommendation Algorithm

In the era of rapid development of the Internet,all kinds of information are also growing explosively.The emergence of recommendation systems allows users to quickly extract content that may be of interest from massive information.However,the features of users and projects extracted by traditional recommendation systems are relatively simple,and the internal relationship between users and products is not enough to be mined,resulting in difficult mining of users’ real preferences,and the recommendation effect is not ideal.Based on the research of the existing recommendation system,aiming at the problem of insufficient extraction of user and item information,this paper improves the existing algorithm to achieve the screening and classification of items and users,and carries out shallow and deep information on user and item information.They are divided and processed respectively to fully extract the shallow and deep information features,and recommend different projects to different users according to their ratings and interests.This paper proposes a recommendation method based on deep noise reduction automatic coder.First,the user and project information needs to be divided into shallow information and deep information,and processed separately.By training the automatic coder separately and using the balance factor to fuse the two coder models,the recommendation of different items for users with different interests is realized.Through the above methods,this paper mainly improves the recommendation accuracy of the current recommendation system and reduces the adverse impact of matrix sparsity.