Vegetable Price Prediction Using Atypical Web-Search Data

Our study focuses on 3 vegetables mainly purchased in Korea; onion, garlic, and dried red pepper. We develop atypical index reflecting consumers’ attention on those vegetables from social network service (SNS) websites and major portal sites such as Google. Specifically, using text mining program, we gather associate web-search data, making simple query data measuring frequency on websites and Term Frequency – Inverse Document Frequency (TF-IDF) considering weights of core keywords on websites. We introduce those asymptotic indexes into the Bayesian structural time series models with climate factors impacting vegetable prices. Results show that the introduction of atypical web-search data can improve vegetable price prediction power compared to pure time-series models without atypical indexes