Research On Inventory Optimization Based On Sentiment Analysis Of E-commerce Reviews And Sales Forecasting

The COVID-19 swept the world from 2020,and consumers’ purchasing habits changed in the epidemic and post-epidemic era."Stay at Home Economic" enabled e-commerce to sustain development.In the context of the further expansion of the number of Internet users and the mobile Internet brought by smart phones,the consumer-centered digital transformation is also continuing.How to use consumer data for data mining to improve supply chain operation has been widely concerned.From the perspective of fully mining consumer preference data,this paper conducts fine-grained sentiment analysis on e-commerce reviews,studies sales forecasting based on sentiment analysis of e-commerce reviews,and optimizes inventory based on sales forecasting to improve and optimize supply chain management.The main work of this paper is as follows:Firstly,in this paper,the fine-grained sentiment analysis of e-commerce reviews is transformed into a multi-label classification problem.From the perspectives of context information,syntax and labels,corresponding text features are extracted and fused based on BERT model,Word2 Vec and modified TF-IDF algorithm respectively.A multi-label classification model based on BERT and modified TF-IDF is proposed.At the same time,the validity of the proposed method on fine-grained sentiment analysis is verified on the real e-commerce review data set of Jing Dong platform.In addition,according to the fine-grained emotion analysis results of real data sets,it is concluded that for fresh goods like fruit,the most concerned attribute is taste,followed by freshness,and the least concerned attribute is customer service attitude.Then,aiming at the sales forecasting problem,it aims to improve the accuracy of the forecasting model by adding the emotional tendency of consumer comments.By building a sentiment index model,this paper converts the discrete results of sentiment analysis into quantifiable values to verify the correlation between sentiment of comments and sales.On this basis,the sentiment index is added into the sales forecasting model as one of the influencing factors.At the same time,inspired by human memory model,we improved the structure of short and long-term memory network,added "enhance gate" and "suppress gate",constructed the enhance-suppress short and long-term memory network.And a sales forecasting model of enhancedsuppress long short-term memory based on fine-grained sentiment analysis is proposed.The effects of sentiment index and enhanced-suppress long short-term memory network on improving the prediction accuracy of the model were verified by comparative experiments.Finally,aiming at the inventory optimization problem,a dynamic(t,R,S)inventory optimization strategy based on sales forecast is proposed on the basis of comprehensive cyclical strategy.The dynamic safety stock is updated based on the sales forecast results,the dynamic safety stock is taken as the reorder point,and the dynamic forecast sales is taken as the maximum inventory level.Finally,taking the out-of-stock risk and inventory cost as evaluation criteria,the paper verifies that improving the efficiency prediction accuracy can effectively reduce the out-of-stock risk and inventory cost,optimize the inventory level and improve the supply chain service level.