Design And Implementation Of Personalized Location Recommendation System Based On Spark

With the advent of the data age,network users need to receive countless different types of information every day,and face the trouble of selecting the required information from the massive information.At this time,the recommendation system came into being.Among the popular big data computing platform,the Spark framework based on memory iteration calculation is more in line with the requirements of the recommendation system.Compared with the traditional Hadoop Map Reduce framework,multiple master nodes in the Spark framework solve the single-node failure problem that is easy to occur in Hadoop,so it has higher real-time computing capability,which can greatly improve the operating efficiency of the recommended system.Based on the analysis of the status of the personalized location recommendation system under the Spark framework,under the premise of understanding the related technologies of system design,the research of this paper is focus on algorithm optimization and system construction.Based on the recommendation requirements of users,this paper proposed recommendation engine architecture combining offline recommendation and online recommendation.The recommendation engine architecture completes the design and implementation of a personalized location recommendation system.In order to improve the data sparse problem and cold start problem in the traditional LBSN-based location recommendation algorithm,this paper proposes a weighted matrix factorization algorithm that combines user preferences and contextual information.Compared with the traditional matrix decomposition algorithm,the weighted matrix decomposition model mitigates the data sparse problem by weighting the elements in the user-check-in matrix.In the implementation process of the recommendation system,the influencing factors of the user sign-in mainly include user preference and context information,wherein the user preference has the greatest impact on the sign-in behavior,The weighted matrix decomposition model used in this paper builds the objective function based on user preference,and then models the impact of context information on the user's sign-in behavior.It constructs the influence matrix of the geographic location on the user's sign-in behavior and the impact matrix of the social network on the user's sign-in behavior,Adding these two matrices to the objective function to redefine the influencing factors of the user's sign-in behavior and predicting the probability of the user going to the target location for sign-in,this method further improves the accuracy and recall rate of the recommendation algorithm.The overall architecture of the system design is divided into four modules: data source,data warehouse,core business and data display.The stable transmission between modules ensures the smooth running of the recommendation system.The recommendation engine is divided into online recommendation and offline according to different requirements to meet the accuracy requirements of the recommended results.Finally,the system is tested on the basis of Gowalla dataset and Foursquare dataset,which proves the high accuracy and high recall rate of the system implemented in this paper.