Design And Implementation Of A News Recommendation System Based On User Interest Mining

With the continuous development of electronic information technology,the speed of news production and dissemination is getting faster and faster.While enjoying the convenience,users are also faced with the trouble of not knowing how to choose in the face of the massive amount of information.News recommendation system can recommend contents that may be of interest to users based on their reading history,which improves users’ experience and is the key to solve the information overload problem.Existing news recommendation methods often model users’interests by the news they have read as the basic unit.Although users’interests can be represented by the news they have read in their history,modeling by news unit will lose the internal connections between news and may lose more fine-grained user interests.In this paper,we propose a fine-grained user interest mining algorithm based on knowledge graphs,using graph attention networks to mine information contained in neighboring entities from knowledge graphs,capturing connections between entities read by users through multi-headed self-attention,and aggregating information of entities read by users to learn fine-grained interest representations of users.In this paper,we combine multi-granularity user interests,modeling user interests at two granularities,news read by users and entities read by users,and fusing coarse and fine granularity interests to make recommendations to users.Experiments on the MIND dataset show that the proposed algorithm is able to learn higher quality user representations and achieve better recommendation results.To address the problem of insufficient positive samples in the training process of news recommendation system,this paper constructs a selfsupervised task by contrast learning,performs data augmentation on the candidate news representations,constructs positive sample pairs by Dropout mechanism,and pushes the distance between the positive and negative sample representations by narrowing the distance of positive sample pairs in space through contrast learning loss.The recommendation task and the self-supervised task are jointly trained,and experiments conducted on MIND data show that the method learns higher quality news representations and improves the effectiveness of recommendations.Combined with the proposed multi-granularity user interest mining algorithm,this paper implements a news recommendation system based on multi-granularity user interest mining.This paper presents the system in detail from requirement analysis,functional design,and system architecture design,and tests the system functionality.The system functions normally,can meet the user’s recommendation needs and has a relatively good performance.