Research And Implementation Of Sequence Recommendation Based On User Interest And Time Information

With the development of science and technology and the popularization of network,network resources are increasing exponentially,and the phenomenon of information overload is becoming more and more serious.People can easily obtain all kinds of resources from the network,but how to provide information that meets the needs of users has become a problem that puzzles people,so the recommendation system came into being.Early recommendation systems mainly used collaborative filtering,matrix decomposition,statistical analysis and other methods.In recent years,due to the rapid development of machine learning,neural network has been widely used in recommendation systems,and the performance of recommendation systems based on neural network is usually better than traditional recommendation methods.We regard the e-commerce recommendation system as a sequential recommendation problem,which aims to predict the next project that users may interact with.However,recent studies on sequence recommendation usually ignore two very important information.First,in the research of recommendation system based on neural network,people usually embed the user’s behavior sequence as a whole.However,a unified user embedding can’t reflect the multiple interests of users in a period of time.Second,most of the existing methods simplify the time information of behavior into behavior sequence,so that the modeling based on cyclic neural network(RNN)can be carried out later.However,in this simple way of sequence recommendation,the key time information are largely ignored.In order to solve the above two important information that is easy to be ignored in the current research,taking the e-commerce platform as the background,this paper proposes a neural network model based on auto encoder and self attention mechanism to capture and utilize users’ multi interest and complex time information,and experiments are carried out on multiple real-world data sets to compare with the advanced sequence recommendation model,It proves the effectiveness of our framework.The main work of this paper is as follows:(1)This paper makes an in-depth study on sequence recommendation,and expounds its development trend and research significance.The advantages and disadvantages of two kinds of sequence recommendation methods(based on traditional recommendation algorithm and based on deep neural network)are summarized and analyzed.(2)Based on the analysis of a large number of existing data,we found a variety of potential interest information and complex time patterns in the behavior sequence of users.In order to better model the user interaction sequence of the recommendation system,a comprehensive deep learning model is proposed,which integrates the multi interest module and complex time module into a unified recommendation system.(3)The model proposed in this paper is compared with several classical and advanced sequence recommendation models in a large number of experiments on multiple real data sets.The model in this paper has achieved the best results in several general indicators,so as to prove the effectiveness of this research in sequence recommendation.(4)A prototype system including sequence recommendation function is designed and implemented.The system realizes the method of personalized recommendation based on user multi interest and complex time information proposed in this paper.