Research On Hybrid Recommendation System Based On Store Location

With the popularity of mobile devices,mobile Internet has developed rapidly.Advertising business,as one of the most important profit models of mobile internet,has been paid more and more attention.With the expansion of the scale of users and the increase of the number of merchants,every consumer faces many choices in different locations.At the same time,merchants are also looking for suitable users.This paper proposes an improved collaborative filtering algorithm based on the location of stores,which can recommend appropriate users based on the location of stores.The main work of this paper is as follows:1.Aiming at the coordinates uploaded by cashiers in stores,an algorithm combining DBSCAN with kmeans is proposed to calculate the coordinates of stores,and the coordinates uploaded during non-busy periods are filtered according to the working time of the transaction table.The algorithm realizes the parallelization of coordinates of multiple stores by spark's RDD partition.Calculate and compare the stability of the calibration algorithm in days,three days and a week respectively,and finally execute the calibration algorithm in units of a week.2.By calculating the store coordinates and using the spark-based distributed DBSCAN algorithm to cluster the stores,it is possible to implement the DBSCAN algorithm for tens of thousands of point sets.The clustering result is the business circle.This paper proposes the criterion for the division of business circles,that is,the promotion degree of the average number of users in the business circle.Three,user recommendation of business circle stores.This paper proposes an improved collaborative filtering algorithm based on the location of stores to recommend users.After calculating the exact coordinates of stores,the distance between stores is calculated by the coordinates of stores.Then the similarity of stores is defined and the nearest neighbor recommendation is made according to the similarity of stores.Compared with the traditional recommendation algorithms,such as cold start,sparse data and narrow application range,the proposed recommendation algorithm based on geographical location is a better idea from the point of view of the algorithm.The proposed recommendation algorithm based on geographical location is better than the traditional collaborative filtering algorithm in terms of recommendation accuracy,recall rate and F1 value.There was a marked improvement.