Article: Integrating sentiment analysis and deep learning for regional economic risk identification and early warning Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.50 pp.52 - 71 Abstract: In this paper, an innovative early warning model integrating news sentiment analysis and deep learning is proposed to address the complexities of regional economic risk identification and early warning. The model extracts spatio-temporal features from macroeconomic indicators and news texts respectively through a dual-channel network structure, and utilises the attention mechanism for dynamic fusion. Experiments based on China's provincial panel data and global database of events, language and tone news data show that this model achieves the harmonic mean of precision and recall of 0.812, which represents a significant improvement of 8.9% over the best-performing benchmark model (XGBoost at 0.745) and 16.3% over the traditional logistic regression model (0.698). Furthermore, this model can identify potential risk areas earlier. These findings provide new methods and decision support technologies for regional economic risk monitoring, which are of great significance to policymakers and financial regulatory authorities. This study is validated on Chinese provincial data, and generalisability to other regions requires further testing. Future work will explore finer spatial granularities and diverse data sources. Inderscience Publishers - linking academia, business and industry through research

Title: Integrating sentiment analysis and deep learning for regional economic risk identification and early warning

Authors: Yize Hong

Addresses: School of Business and Management, Jilin University, Changchun, 130012, China

Abstract: In this paper, an innovative early warning model integrating news sentiment analysis and deep learning is proposed to address the complexities of regional economic risk identification and early warning. The model extracts spatio-temporal features from macroeconomic indicators and news texts respectively through a dual-channel network structure, and utilises the attention mechanism for dynamic fusion. Experiments based on China's provincial panel data and global database of events, language and tone news data show that this model achieves the harmonic mean of precision and recall of 0.812, which represents a significant improvement of 8.9% over the best-performing benchmark model (XGBoost at 0.745) and 16.3% over the traditional logistic regression model (0.698). Furthermore, this model can identify potential risk areas earlier. These findings provide new methods and decision support technologies for regional economic risk monitoring, which are of great significance to policymakers and financial regulatory authorities. This study is validated on Chinese provincial data, and generalisability to other regions requires further testing. Future work will explore finer spatial granularities and diverse data sources.

Keywords: regional economic risk early warning; sentiment analysis; deep learning; attention mechanisms; multi-source data fusion.

DOI: 10.1504/IJICT.2025.151071

International Journal of Information and Communication Technology, 2025 Vol.26 No.50, pp.52 - 71

Received: 09 Oct 2025
Accepted: 11 Nov 2025
Published online: 12 Jan 2026 *

Title: Integrating sentiment analysis and deep learning for regional economic risk identification and early warning

Keep up-to-date