| With the continuous improvement of China’s retail industry’s GDP and residents’ consumption levels,the development of chain convenience store companies is changing rapidly.While facing development opportunities,it is also facing difficulties in the operation process.A store has a limited business area,usually no more than 100 square meters,and the number of products that can be placed on the shelf is extremely limited.The store can radiate a customer group within a radius of 500 meters.The customer group served by different stores may have very different preferences,and similar products are in the market.There is also the problem of new and old replacements,so convenience stores need to periodically update in-store products,remove products with poor sales,and introduce potential best-selling products to increase store profits and maintain business survival.The experiment proves that the method in this paper effectively improves the store profit improvement level after the commodity update,and improves the stability of the profit level change,and reduces the occurrence of extreme cases of profit decline.The updating of the products placed in the convenience store has been updated by the store manager based on experience and prediction,combined with the group’s business strategy.This method lacks stability and continuity.This article designs and implements a deep learning-based product recommendation algorithm for the periodic update of convenience store products.The specific work includes the following points:Firstly,in order to realize the product recommendation algorithm based on deep learning,the application status of computer technology in the chain convenience store industry was researched,and the similar store analysis was determined—the candidate product sales forecast—the mathematical model of the store profit maximization solution.The main technical route,and carried out many aspects of research and in-depth research on various fields of technology.Secondly,as the basis of the product recommendation algorithm,the store sales records of each chain store in the past two years and the group product master file were analyzed and organized,and feature vectors expressing the characteristics of the store were constructed in different subspaces to provide similar store analysis later.Calculation basis.Thirdly,similar store analysing.Clustering multiple stores in a chain convenience store is an unsupervised learning process.The number of clusters affects the clustering results.Through the multi-objective optimization method,optimization is performed among multiple cluster evaluation indicators.It can automatically find the appropriate number of clusters according to the sales status of all stores in the current period,and merge the adaptive clustering results of different subspaces separately to obtain similar stores under comprehensive consideration,which is more reasonable.Fourthly,to predict the sales volume of the next period of candidate products,based on the daily sales time series,introduce external information as covariates,and use longand short-term memory neural network models for prediction,which can effectively improve the prediction accuracy and use the adaptive evolution strategy independently adjusts the time series of the products in each store,and obtains high-precision sales forecast results.Finally,to establish a mathematical model for product selection based on business goals.The objective function is to maximize the profit of the target store.The decision variables are all the products in the target store and the Acacia store in the current period.The business needs and the actual limitations are converted into constraints.The result of the solution is the result of the recommendation for the product of the target store.Experiments show that the product recommendation result can increase the profit level of the store.Experiments show that the product recommendation algorithm can theoretically effectively increase the profit level of the store.This article takes the operation data of a chain convenience store in Beijing as a research object.The company’s 300 stores have12 product updates in 2019(the beginning of each month),and the average operating profit increase of all stores in the company is 1.17% per month.After the proposed product learning algorithm based on deep learning,the theoretical average profit increase reached 1.67%,an increase of 42.74%.In addition,the profit improvement level of each period is more stable,reducing the degree of lubricating slope and the frequency of occurrence.The minimum value of the average profit increase value of each store for commodity updates has been increased from-0.17% to-0.08%,and the number of negative profit increases has been reduced from 4 3 times. |
