Design And Implementation Of Dance Micro-video Recommendation System Based On Deep Learning

With the popularity of the Internet,it has become an essential way to enjoy and learn dance online.However,due to the rapid development of dance video websites,the problem of information overload is becoming increasingly prominent.Therefore,in this case,using the recommendation system for dance recommendation is an effective solution.As an effective tool,deep learning technology can solve complex feature extraction and feature crossing problems in traditional recommendation models.More and more studies begin to pay attention to the important role of recall stage,so as to improve the efficiency of recommendation system.Therefore,the recommendation system is divided into two stages: recall and sorting,so as to recall as many items related to users as possible in a short time.In addition,a complex model based on deep learning is adopted in the ranking layer to improve the accuracy of the model,so as to achieve better recommendation effect.The research content mainly includes the following four aspects:(1)Collect and process the data needed for the project.This topic collects a total of three types of data,including the detailed information of the dance video,the user’s information and the user’s interaction information.The specific situation of the three types of data and the specific steps of preprocessing are introduced.(2)Multi-way recall model construction.By quickly screening the full number of dance micro videos,the data set is reduced to thousands of levels,and the multi-way recall algorithm model is merged to obtain the final candidate set.The multi-way recall algorithm model uses Word2 vec model to extract text feature vectors,and uses itembased collaborative filtering algorithm,factorization machine model,two tower model based on deep learning and You Tube DNN model to recommend recall.These four different algorithm models generate recall candidate sets and form the final recall list for subsequent models.It can effectively solve the problem of low recall rate caused by single traditional recall method.(3)Construction of ranking model.This paper proposes to use the large-scale pretrained model Bert for text feature extraction as a supplementary feature input of the model,and combine it with the XDeepFM model based on deep learning to reduce the difficulty of user and item feature acquisition.At the same time,it is compared with other mainstream recommendation models under this dataset.The results show that the recommendation effect of the proposed model can effectively use the characteristics of dance and user,and improve the recommendation effect.(4)Implement the dance micro video recommendation system.Combined with the actual needs of users,the multi-channel recall model and the fusion Bert and XDeepFM model are applied to the dance micro video recommendation system.Based on the system architecture,the recommendation system is developed to realize the functions designed by the recommendation system,and the recommended dance micro videos selected by the model are displayed to users.