Design And Implementation Of A Fast-moving Consumer Goods Platform Based On Multi-dimensional Recommendation Algorithm

With the development of the e-commerce field and the changes in consumer shopping patterns,there is a growing demand for people to buy fast-moving consumer goods online.Fast-moving consumer goods called FMCG for short,refer to those consumer goods with short lifespans and fast product consumption.Compared with industries such as durable consumer goods,FMCG is a unique field.FMCG shows different characteristics in procurement,marketing,and storage.Affected by the epidemic,offline FMCG merchants have suffered a certain impact when selling goods.More and more consumers are more inclined to buy goods online and deliver them to their homes by express delivery.In the case of fierce market competition,it is very important for FMCG merchants to open up online and offline business channels.This thesis designs and implements a fast-moving consumer goods platform based on a multi-dimensional recommendation algorithm.The platform provides online and offline procurement,sales,storage and other functions for FMCG merchants to do business.The platform helps merchants manage goods,daily purchases and sales,manage warehouses,manage funds,and promote marketing.In the FMCG platform,merchants can adapt to different FMCG industries to achieve refined management of commodities.The platform supports the use of multiple devices to quickly issue orders in online malls or offline stores,check commodities in real time,adjust prices,manage inventory,formulate promotional activities,send logistics,exchange online and offline data,and realize retail and wholesale management of commodities.The platform supports merchants to formulate purchase orders at any time to help merchants meet the needs of daily purchases and purchases.At the same time,the platform supports multi-channel simultaneous selling of goods for merchants,provides merchants with a variety of marketing methods,realizes multi-platform collaborative sales,and helps merchants increase sales.In addition,the platform supports merchants to view commodities,inventories,and sales data at anytime and anywhere,push business bills in real time,and help merchants operate and manage stores.The modules that the author mainly participated in and implemented include novice guidance,basic information management,procurement management,retail management and wholesale management.The front-end of the FMCG platform is developed using the React framework combined with the low-code platform and Node.js for pages development.The back-end adopts the microservice architecture.The My SQL database is used for data storage to store data and Redis is used to store cached data.In current market,there are many types and quantities of commodities,the frequency of updates is fast,and the personalized needs of customers are becoming more and more unique.Faced with the massive variety and quantity of commodities,how to accurately find out the commodities that consumers want to buy,and which commodities to increase their efforts to promote during sales to obtain greater benefits,are problems that businesses need to consider and solve.In response to the above problems,this thesis adds a multi-dimensional recommendation algorithm on the basis of the FMCG platform,and the results of the recommendation algorithm will be applied to the retail mall of the retail management module.The recommendation algorithm will start from multiple dimensions such as users,commodities,and scores to intelligently recommend commodities for consumers and help merchants conduct precise marketing.The recommendation algorithm used in this thesis includes two aspects: explicit feedback and implicit feedback.(1)In explicit feedback,this thesis proposes a recommendation algorithm based on an improved lightweight graph convolutional network.In order to solve the problem of data sparsity,after creating the adjacency matrix the improved lightweight graph convolutional network adds the scoring matrix after SVD matrix decomposition for feature fusion.At the same time,the improved lightweight graph convolution network will use the Sentence-BERT model with better effect to embed the text field,obtain the text information embedding vector of the commodity,and then fuse its features with the commodity embedding matrix obtained by graph convolution operation to enrich the embedding matrix of the commodity.(2)In implicit feedback,this thesis proposes a recommendation algorithm based on an improved autoencoder network.The improved autoencoder algorithm adds two feature fusions and performs two trainings.The first time is to perform feature fusion between the predicted score matrix after autoencoder training and the predicted score matrix obtained by the score-based collaborative filtering algorithm.Then,the fused matrix and the text information embedding matrix trained by the Sentence-BERT model are used for feature fusion,and finally the fused matrix is sent to the autoencoder algorithm again for training.After the system test,the FMCG platform was launched as scheduled.After verification by merchants,the FMCG platform can solve problems such as slow billing,inability to issue orders when leaving the store,and confusion in inventory management,improving the billing efficiency and inventory management capabilities of merchants.At the same time,the improved algorithm proposed in this thesis has achieved better results in terms of MSE,precision rate,recall rate and other indicators,which is helpful to improve the marketing effect of commodities.