Scholarlink

<p>In real-world financial markets, a company’s stock performance is often influenced by its competitors, partners, and other connected firms. However, traditional sentiment-based approaches typically focus only on the target company’s sentiment, overlooking the broader network of influence. To address this limitation, this paper proposes a Dual Transformer architecture that integrates sentiment signals from connected companies to enhance prediction accuracy. This approach aims to bridge a critical gap in financial sentiment analysis by capturing inter-company relationships and their impact on stock price movements. The Dual Transformer comprises two transformer structures, the Enhancement Transformer and the Forecast Transformer. The Enhancement Transformer is used to enhance the correlation strengths between related companies, while the Forecast Transformer is used for stock price forecasting. Eight companies from various industries and financial markets are selected for analysis. The model utilizes the polarity scores of these related companies, combined with historical closing prices from 2015 to 2024, to forecast the next day’s closing price. Prediction performance is evaluated using the mean squared error (MSE). Experimental results demonstrate that incorporating the news sentiment of related companies improves the prediction of the target company’s stock price when using the proposed Dual Transformer model. In addition, the Dual Transformer model can outperform existing models such as Temporal Fusion Transformer, N-Beats, Informer, and LSTM by achieving consistently lower MSE values across all eight companies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel sentiment correlation-based method with dual transformer model for stock price prediction

  • 期刊论文

  • 发表日期：

    2025年12月2日

查看全文
  • Qizhao Chen,
  • Hiroaki Kawashima

摘要

In real-world financial markets, a company’s stock performance is often influenced by its competitors, partners, and other connected firms. However, traditional sentiment-based approaches typically focus only on the target company’s sentiment, overlooking the broader network of influence. To address this limitation, this paper proposes a Dual Transformer architecture that integrates sentiment signals from connected companies to enhance prediction accuracy. This approach aims to bridge a critical gap in financial sentiment analysis by capturing inter-company relationships and their impact on stock price movements. The Dual Transformer comprises two transformer structures, the Enhancement Transformer and the Forecast Transformer. The Enhancement Transformer is used to enhance the correlation strengths between related companies, while the Forecast Transformer is used for stock price forecasting. Eight companies from various industries and financial markets are selected for analysis. The model utilizes the polarity scores of these related companies, combined with historical closing prices from 2015 to 2024, to forecast the next day’s closing price. Prediction performance is evaluated using the mean squared error (MSE). Experimental results demonstrate that incorporating the news sentiment of related companies improves the prediction of the target company’s stock price when using the proposed Dual Transformer model. In addition, the Dual Transformer model can outperform existing models such as Temporal Fusion Transformer, N-Beats, Informer, and LSTM by achieving consistently lower MSE values across all eight companies.