A multisource data‐driven combined forecasting model based on internet search keyword screening method for interval soybean futures price

Abstract Accurate soybean futures price prediction is critical to related agricultural production, warehousing, and trading. Interval forecasting can avoid the loss of fluctuation information and evaluate the uncertainty of futures prices. However, most previous studies only consider the single‐type auxiliary variable, which will cause the deficiency of valued information. Moreover, the research concentrating on internet search index ignores the search habits of investors, resulting in subjectivity in keyword selection. Therefore, a novel multisource data‐driven combined forecasting model is proposed that consists of four parts: unstructured data processing, interval multi‐scale decomposition, interval combination forecasting, and model evaluation. First, sentiment analysis technology is used to convert news text into sentiment scores. The internet search keyword screening method based on latent Dirichlet allocation is then constructed to achieve the quantification of investor attention. Second, a decomposition method is applied to decompose the original interval‐valued series into finite more stationary components. Third, interval prediction results are obtained by machine learning‐based multiple predictors. Finally, the model evaluation module comprising error evaluation indicators and comparison experiments is presented to verify the effectiveness. The experimental results show that the proposed model has higher prediction accuracy, which indicates that multisource data and designed keyword screening methods can enhance the forecasting performance of the model.