Application And Research Of Hybrid Recommendation Algorithm In Vending Machine Commodity Sales

In real life,while the vending machines bring the convenience of unmanned vending to the suppliers,they also bring massive amount of data which leads to information overload.How to use the overloaded information to improve the efficiency for the merchandisers is the fundamental goal of the recommendation systems.However,the current vending machine product recommendation systems are all for users,and there are very few product recommendation algorithms specifically for vending machines.The issue with vending machine product replenishment and cold start issue with newly launched vending machine have always been the pain point and popular research topic in the industry.To this end,this thesis explores and implements a product recommendation framework for vending machines based on hybrid recommendation algorithm.The contents of this thesis are as follows:(1)A product recommendation model for vending machines based on a hybrid recommendation algorithm is developed in addressing the problem of low recall rate and poor recommendation effect of hot-selling products.The NeuralCF model and the Wide&Deep model are combined through the Blending method to generate a hybrid recommendation model,realizing the complementary advantages of the two algorithms.The experimental results show that the AUC(Area Under Curve)value is increased by about 6.7% and 3.0% compared with the NeuralCF model and the Wide&Deep model,and the accuracy rate is increased by about 2%-7%.(2)A linear replenishment model is designed and implemented to address the problem that the type and quantity of goods carried by vending machines are uncertain and the replenishment loss is relatively large during the replenishment process of vending machines.Firstly,determine the replenishment type based on the above mixed product recommendation results,then use the mixed prediction algorithm to forecast the sales of the recommended products,and finally use the linear replenishment model to generate the replenishment list according to the predicted sales volume.The experimental results show that the replenishment model can reduce the total replenishment loss effectively under the condition of meeting the normal sales volume of vending machines,and the total replenishment loss is reduced by about 7.64%.(3)In view of the cold start problem of the newly launched vending machine,combined with the geographical location of the vending machine,whether there is a dedicated venue,color,capacity and other information,the KNN algorithm is used to calculate the vending machine that is more similar to the newly launched vending machine,and obtain the initial commodity list for the new vending machine according to the commodity list in this type of vending machines.The online operation results show that compared with the unified initial list without the nearest neighbor method,this method increases the sales of vending machines by about 13.4%.(4)The thesis designed and implemented a Web-based vending machine product recommendation system.It has been accomplished functions such as the management of the vending machine and the recommendation for the product.Field operation results show that the system can increase the average sales volume of vending machines by about 4.3%,which verifies the effectiveness of the constructed vending machine commodity recommendation model and the model to solve the cold start problem.