Research On Recommendation System Model Based On Deep Neural Network

With the outstanding performance of deep learning in the fields of computer vision,object recognition,speech recognition,etc.,deep learning has received extensive attention and is gradually applied to other fields.The recommendation system is a very efficient information processing tool in the "big data" era.When people face huge and redundant information and do not know how to search or obtain the information they want,the recommendation system can predict the information they want.And show it to people.The recommendation system model based on deep learning can better mine value information,thereby improving the recommendation accuracy and work efficiency of the recommendation system model.However,there are still problems in applying deep learning technology to the recommendation system today.In this paper,we propose two deep learning recommendation models,the graph neural network news recommendation system IGNR based on long and short-term interest points and the CDHMF recommendation model based on hybrid deep matrix decomposition.(1)For the IGNR model,we construct a heterogeneous user-news-topic graph to model user-project interactions,thereby reducing the sparseness of user-project interactions.Then,it applies graph convolutional networks to learn users and news embeddings with higherorder information encoded by propagating embeddings on graphs.(2)For the CDHMF model,we not only use explicit ratings,but also use implicit feedback as input to learn the potential feature space of users and projects,and integrate the auxiliary factor vector extracted from the project content text by the RNN encoder into the Neural network architecture to optimize the presentation of the project.Experimental results on real data sets show that the two models we proposed are significantly better than the latest methods in their respective fields.