Big Data Time Series Prediction Based on Feature-Enhanced Transformer Model

Runqing Tang, Yawen Li, Zhe Xue, Junping Du, Guanhua Ye

Abstract: As the digital economy gradually becomes a new engine for economic development in China, it is crucial to promote its advancement through strategic digital transformation by enterprises. By delving into corporate innovation needs, user demands for new products, and technological resources, we analyze and predict the evolving relationships and trends in corporate innovation requirements. This approach assists enterprises in understanding market demands and competitive landscapes, ultimately enhancing their competitiveness and innovation capabilities. We downloaded datasets related to corporate innovation from the CSMAR official website, which primarily include basic corporate information, patent details, and innovation index data. The datasets underwent processes such as data cleaning, integration, feature selection, and loading to improve data quality further. A sliding window algorithm was introduced to meet the model’s data loading requirements, resulting in a standardized corporate innovation datasets that clarifies related tasks based on the processed data. We propose a feature-enhanced Transformer model for the time series prediction of innovation indices, achieving the expected accuracy and enabling effective forecasting. This model provides enterprises with more accurate and comprehensive innovation prediction results, offering stronger support for the formulation of their innovation strategies.

External IDs:dblp:conf/bigcomp/Tang0X0Y25