Predicting sales by online searching data keywords based on text mining: Evidence from the Chinese automobile market

Abstract

Online searching data reflects consumers' real footprints in information collection and purchase decision-making processes, which is greatly valued in understanding their needs. This paper which is at the background of China's automobile market, studies the relationship between online searching data and automobile sales through approaches that differ from existing research to extract keywords. First the online searching data keywords are determined, primarily by using text-mining technology to extract them, and specifically: i) Jieba was used to tokenize crawled automotive forum posts' text into segmented words; ii) All word-segmented Chinese corpus were segmented into word vector space by Word2vec model; and iii) Similar keywords were discovered by calculating the word vector's similarity indexes. A fixed effect model was then built based on 108 months of long panel data. Finally, combing with panel vector autoregressive model (PVAR), we used rolling window to predict Chinese automobile sales from January to December 2015.The empirical results demonstrate that: a long equilibrium exists between online searching data and automobile sales; our regression model can explain 76% of the variance. The holdout analysis suggests that online searching data can be of substantial use in forecasting Chinese automobile sales.