Text Sentiment Analysis Assists The Forecast Of Beijing's Total Retail Sales Of Consumer Goods

The total retail sales of consumer goods are of great value for the formulation of economic policies and scientific research.A large number of researchers have established prediction models based on time series correlation theory and econometrics.In recent years,many scholars have tried to use non-economic data that can be collected in real time to predict the total retail sales of consumer goods.Such as the search index related to consumption by search engines,consumer confidence index,etc.The addition of these data has made the prediction accuracy of total retail sales of social consumer goods significantly improved.The total retail sales of consumer goods in Beijing are affected by many factors.It is difficult to analyze through traditional methods.The addition of affective factors can significantly improve the prediction effect of traditional prediction models.Therefore,this article selects the total retail sales of consumer goods in Beijing as the research object,and uses the technology of text sentiment analysis to establish the sentiment factor index to assist in the prediction of the total retail sales of consumer goods in Beijing.This article firstly makes a detailed analysis of the total retail sales of consumer goods in Beijing,and lays a theoretical foundation for the addition of textual sentiment factors.This article uses focused crawler technology to crawl news from news sites for sentiment analysis.Sentiment factors based on sentiment dictionary and deep learning are used to establish monthly sentiment factors for sentence granularity and chapter-level granularity,respectively.By establishing a vector autoregressive model and Granger causality analysis,the rationality of introducing two emotional factors into the prediction model is demonstrated.Establish the SARIMA model based on the time series theory and the LSTM baseline model based on the neural network for the total retail sales of consumer goods in Beijing,and then introduce the two granular sentiment factors into the SARIMA model and the LSTM model respectively,and compare the model with the sentiment factor and the baseline The prediction accuracy of the model.It has been demonstrated that affective factors have auxiliary effects on time seriesbased models and neural network models,and sentence-level affective factors have better auxiliary effects because of the finer granularity.Finally,this paper merges the LSTM model and the SARIMAX model to establish the LSTM-SARIMAX model,which improves the stability and prediction accuracy of the model.Compared with the single-model best performing LSTMX1 model,the prediction accuracy is improved by 8.39%.