Consumer Confidence Index Analysis Based On Web Searching Data

In the big data era, massive network data contains all kinds of consumer behavior, the data in real time to reflect changes in the socio-economic conditions and psychological expectations of consumers instantly. How to extract the important information from these massive data to analyze the macro economic phenomenon is a hot spot in the future research direction. There are some problems in Consumer Confidence Index : long released time, low timeliness, narrow coverage. Under this background, this paper combines innovatively the network data to forecast the Consumer Confidence Index, which is an important supplement to the formulation of the Consumer Confidence Index.Based on data mining method and Internet search data, this paper builds an analytical framework of Consumer Confidence Index prediction model. Using time difference analysis, this paper selects search keywords which have a strong correlation with Consumer Confidence Index and are ahead of the changes of the Consumer Confidence Index from 40 web search keywords.As the Consumer Confidence Index for the monthly data, and the training sample is less, the direct establishment of the model is easy to over fitting. In this paper, stepwise regression, lasso algorithm, ridge regression and principal component analysis are used to reduce the dimension of the selected search keywords, comparing and analyzing the effect of different dimensionality reduction method. On this basis, three data mining methods, support vector machine, decision tree and GBDT are use to forecast Consumer Confidence Index. The empirical proofs that the selected keywords can reflect the consumer behavior habits, and that the model prediction results are a month earlier than the official data which are able to effectively predict the turning point of the change of Consumer Confidence Index.