Research On Recommendation Algorithm Based On Recurrent Neural Network

In recent years,with the rapid development of internet technology,the data on the network was explosive growth.Massive data on the one hand makes it easier for people to get rich information,on the other hand people have to spend a lot of energy and time to search for useful information,information overload problem more and more serious.In the face of massive data,the traditional search engine has been unable to meet the needs of users,recommendation system has become the new darling of the times.Recommendation system capture the user’s interest by analyzing user data,recommend to user the information or items that he interested.Collaborative filtering is a popular recommendation algorithm that predicts or recommends based on the user’s rating or behavior.In real life,the common recommendation system(such as the recommendation system of the movie website)calculates the vacancy elements in the rating matrix based on the ratings of the purchased or viewed movies by all the users,and recommend high rated items to users.However,when predicting the user will consume which specific goods next time,the traditional collaborative filtering algorithm performance is not good,can not capture the evolution of user interest or context-related interests.In this dissertation,provide a more accurate recommendation by modeling the user’s consumption history.First,in this dissertation proposal a model framework Rating-RNN based on the Recurrent Neural Network(RNN)recommendation algorithm,which introduces the product rating information,and makes a more accurate forecast of which product the user will consume.The local sensitive hash is used to process the data,and the basic recurrent neural network is modified for the problem of movie recommendation.By increasing the Input Gate to introduce the user rating information;adding the Forget Gate to determine how much information is discarded from the previous memory;for the user data sequence length ranging from the regular order;through the output layer hierarchical design to improve the output layer computing efficiency.At the same time,a model framework,Category-RNN,based on the recommendation algorithm of the recurrent neural network,which introduces the information of the category information,is proposed to predict the categories of items that users will consume.Finally,the Rating-RNN is merged with the Category-RNN output layer to form a Mixing-RNN that can be trained in parallel.In the commonly used large-scale public data set MovieLens on the test results show that,compared with the traditional recommendation algorithm for users of short-term commodity consumption forecast,this method has achieved better results.