A Prediction On The Amount Of Automotive Sales Based On Sentiment Analysis Of Reviews

With continuous development of China’s economy,China has now become the largest automobile production and sales country in the world.The automobile industry has also become one of the important factors driving China’s economic development.Therefore,the accurate forecast of car sales will help manufacturers to formulate more reasonable production and sale plans and effectively prevent a series of problems arising from excess capacity.At present,most researches on sale forecast of car are based on historical sales data and use time series for forecasting.Most of the forecasted targets are for the demand of all cars in a region or the entire market,but lack of the forecast method for single model or brand.These studies ingnored the vast amount of car review information on the Internet and lacked in-depth excavation of these network big data.This thesis proposes a method for constructing a forecast model of cars’ sales,for a single model,which uses sentiment analysis techniques to extract emotion information from review data.Then,constructing a single vehicle sales forecast model named Sentimental Auto-regressive model based on comment sentiment and historical sales data.The experiment shows that the sales forecast model has a good effect for a single vehicle comparied with the Auto-regressive model.The reviews are classfied into comfort,power,safety,economy,manipulation and service in this study,by analying the performance of these six emotional values in forecasting to judge what consumers are focusing on the car and provide some basis for cars’ design of auto company.As a high value commodity,automobiles have many factors influencing their sales,including the factors of the automobile itself,and some external factors.Traditional car sales forecasting research only considers one or two factors.This thesis uses Granger causality test to determine the four factors of per capita GDP,highway mileage,steel production and car comment emotion,and then determine the size of the impact through the analysis of gray correlation.Finally,based on these several influencing factors,a regression analysis was conducted to construct a vehicle sales forecasting model.Experiments show that the average relative error of the model in the prediction of car sales is small,and the prediction accuracy is greatly improved compared with the traditional prediction methods.In this thesis,two prediction models are proposed from different perspectives,which can be used in automobile sales forecasting,and can provide support for automobile manufacturers to formulate production and sales plans.