Analysis And Prediction Of Sales Performance Considering Sentiment On Different Platforms

Predicting customer demand is an important part of supply chain management,because it helps to avoid product backlogs or out-of-stock problems.For a booming e-commerce supply chain,accurate demand is generally reflected in sales forecasts.Fully consider multiple factors online,including product prices,reviews,etc.Online consumers can easily obtain a large amount of product information on e-commerce platforms or other channels on the Internet before making purchase decisions.In particular,UGC,represented by user reviews,makes customers not only rely on marketing staff to make promotions Insurance purchase decision.At the same time,for online sellers or e-commerce platforms like Jingdong Mall,making full use of online information,especially online reviews,can analyze the buyer's emotions and attitudes as well as the future sales of products to effectively manage the supply chain.However,the content of customer reviews is large and complex,and exists in the form of text.It is cumbersome to handle and sometimes ignored by researchers.Instead,it focuses on factors such as ratings,number of reviews,and length of reviews.Obviously,If the text sentiment of this part of the review is incorporated into the usual framework of considerations,we can get a clearer understanding of the various factors that predict sales.In addition,review information from third-party platforms will also appear in the consumer's field of vision,and may also have a certain influence on consumer shopping decisions.Therefore,based on the sales influencing factors proposed by previous scholars,this article adds the interaction between the emotional indicators and variables of e-commerce reviews and third-party reviews as prediction indicators,and compares the changes in sales forecasting models to facilitate the discussion of the impact of emotional factors Level,and compare the predictive power and importance of other impact factors,hoping to help consumers and sellers obtain more effective sales information and market feedback;another purpose of this study is to prove various predictive modeling techniques,including linear regression models,Artificial neural network and support vector regression model,in the prediction of the effectiveness of commodity sales on e-commerce platforms,but also reflects the stability of variables in different models.According to the research background of this article,Jingdong Mall and Weibo were selected as two platforms for data collection,and the collection time was one month.Through a review,this paper finally selected 12 independent variables and three prediction methods,namely linear regression,BP neural network and support vector regression.In addition,the sentiment calculation rules for customer reviews are based on the sentiment dictionary method,while Weibo 's sentiment on reviews The indicator selects the relevant data of the microhotspot platform.From the perspective of the sales prediction performance of the three models,the artificial neural network performs best,followed by the linear regression model,and the support vector regression has a poor effect,which also fully reflects the characteristics of the neural network model's strong learning / adaptability.Through the experimental comparison of the three models,it is found that the number of comments is considered to be an important predictor variable,and the e-commerce comment sentiment parameters are not considered as important predictors in the three models,but the sentiment scores in the three models are positive * Poor review score,bad review sentiment score * The regression coefficient of positive reviews shows that it is an important predictor of sales.Secondly,the third-party platform's Weibo popularity and Weibo sentiment are the predictors of the three models,and according to the performance of the three models,the ecommerce platform's comment sentiment is more important than the third-party Weibo platform.Third,poorly rated store responses,membership status,and time to market will not significantly affect sales.Although the final price and the amount of promotion are not significant enough,price * favorable comment,promotion amount * bad rating are considered as important predictors of the three models.The results of this study have certain implications for consumers and store sellers.Consumers can use the results of this article to collect more detailed and effective product information to reduce online shopping risks;stores can fully understand product feedback and achieve more accurate sales forecasts And marketing methods.