Desigh Of Recommendation System Based On User Interest Transfer Model

Traditional recommender systems assume that if users are interested in some commodities in the past period of time,then the similar users may have the same interest.However,the user characteristics will not be stable,and the original features will be transferred over time.Considering that the recommendation system based on collaborative filtering ignored the influence of user interest transfer,therefore,based on the user interest transfer path data,the recommendation based on behavior sequence modeling can improve the quality of recommendation.This paper focuses on three parts.In the first part,this paper applied data mining techniques on user behavior data to mine the potential interest features of each user,then the user interest feature transfer mode is modeled by combining user preferences,data category attributes and user behavior events.In the second part,this paper designed a recommendation system based on user interest transfer model through four steps:demand analysis,function overview,architecture design and database design.In the third part,based on the achievements of the first two parts,this paper predicts the evolution of user interest in the future,and based on the results of model prediction,implemented the recommendation system based on user interest transfer model.In this paper,through the research on the current situation and shortcomings of common recommendation models,the significance of user interest transfer model in recommendation system is clarified.The data preprocessing process extracted the interest transfer path and characteristics for each user.After that,this paper improves Alibaba's deep interest transfer network algorithm and proposes Conv-DIEN model.The results on Amazon books dataset shows that the proposed algorithm is significantly better than the benchmark algorithms,and achieves better modeling effect.Next,the recommendation system based on user interest transfer model is designed.Combined with the existing requirements,the main functions of the system are discussed,and the core process of the recommendation engine is planned,including the acquisition of user historical interest transfer data,the real-time modeling of user interest,and the real-time recommendation process of online personalized accurate recommendation.In the aspect of system implementation,this paper starts from the system requirements,defines the interface and selects the technology,completes the design and implementation of recommender system code based on tensorflow,and deploys the recommender system to the cloud.Compared with the characteristics of other similar systems,the scalability and performance of the system proposed in this paper are simply analyzed.Finally,combined with the development trend of Internet recommendation algorithm,the significance of recommendation system based on user interest transfer is concluded.