Design And Implementation Of Store Recommendation System Based On Deep Learning

In this era of information overload,both information consumers and information producers have encountered huge challenges: for information consumers,it is becoming more and more laborious to find their own points of interest from massive information,As an information producer,they often need to take more account of how to find the potential interests of consumers.Due to the popularity of online consumption,means to attract users’ consumption have changed from promotion to word of mouth marketing.Ratings data have become more valuable.In the field of local life services,after fully mining ratings data,the store recommendation system can provide personalized recommendations to users in time to meet the challenges of information overload.With the emergence of various distributed computing platforms,it provides opportunities for the further development of personalized recommendation algorithms.Among them,spark platform is more suitable for big data scenarios with high real-time requirements,data mining and machine learning scenarios with complex computing models because of its usability,high efficiency and speediness.In recent years,deep learning has been widely used in the field of personalized recommendation.Aided by tensorflow platform,it can quickly build deep learning models,complete parallel training and deploy online.Under the technical background of the new era,the main contents of this thesis are as follows.1)This paper constructs a deep learning recall model based on EGES.Based on the conventional graph embedding algorithm,this paper introduces the side information of store including the city of store,the category of store and the region of store generated by Geo Hash algorithm.The deep walk algorithm is used to model all kinds of attributes of stores and get the matching embedding vector.After that,we can further integrate them for more accurate embedding vectors.with the help of embedding vectors’ relatively strong expression ability,it can quickly recall candidates from a large number of candidate sets.2)This paper constructs a deep learning ranking model based on DIN.Attention mechanism is imported to calculate the weight summation of users’ past behavior features so as to improve the recommendation accuracy of the recommendation system at the ranking layer.3)Finally,we have built a personalized store recommendation system based on spark and tensorflow platform.The algorithm proposed in this paper is used to build recommendation module of the system,and the real-time streaming data is processed through spark streaming and kafka so as to complete the design and implementation of different business recommendation scenarios.Through various tests on the system,it is ensured that it can meet the demands in use of different scenarios on the basis of stable operation.