Research on carbon trading price prediction based on network news data fusion

WANG Rui; LIN Yujian

doi:10.3969/j.issn.1002-9605.2025.12.005

Abstract：Under the global climate governance and China's"dual carbon" strategy,accurate prediction of carbon trading prices is crucial for risk management and policy formulation.To address the limitations of traditional models in capturing unstructured information such as market sentiment and policy expectations,this study integrates online news text with structured time-series data to construct a multimodal carbon price prediction framework.Using the Shenzhen carbon market as a case study,we integrated 10 types of structured daily indicators and 1 893 carbon market-related online news articles from January 2018 to May 2025.Text mining techniques were employed to extract five topics and sentiment features,which were combined with time-series data to establish a cross-modal fusion mechanism.Comparative experiments were conducted using four deep learning models including long short-term memory(LSTM),bidirectional long short-term memory(BiLSTM),gated recurrent unit(GRU),and convolutional neural network(CNN).The findings indicate that multimodal data collaboration significantly enhances prediction performance by effectively capturing unstructured information like policy signals and market sentiment.The GRU model demonstrated the best performance,with its gating mechanism showing notable advantages in balancing computational efficiency and long-term dependency modeling.This study validates the value of unstructured online text data in carbon price prediction,and the proposed cross-modal fusion method provides a new paradigm for carbon price forecasting in complex market environments.Future research could extend this approach to multiple regional markets and incorporate multimodal data such as images and videos to enhance decision-support capabilities.