Emotion measurement method based on improved temporal global principal component analysis and Seq2Seq model

This study focuses on measuring investor sentiment and its impact on stock market behavior. By integrating an improved time-series global principal component analysis and Seq2Seq model, the aim is to address the challenges in quantifying investor sentiment and predicting market behavior. In this paper, we use a modified time series global principal component analysis approach to process market data, extract key features, and combine it with the Seq2Seq model to predict the correlation between sentiment indices and market trading behaviors (e.g., turnover rate and trading volume). This study ensured the effectiveness and predictive accuracy of the model through rigorous feature selection and model training, including the use of Kendall correlation analysis and Granger causality test. The model training and validation adopted a 7:2:1 training validation testing dataset partitioning ratio to enhance the model's generalization ability and prediction accuracy. The experimental results showed that the accuracy of the Seq2Seq model in predicting the correlation between trading volume and investor sentiment reached 0.8713 and 0.9158 before and after adding sentiment index features, respectively, verifying the effectiveness and reliability of the Seq2Seq model in measuring investor sentiment and predicting market behavior. This study provides a practical and feasible solution for automated investor sentiment monitoring, and lays the foundation for further exploration of multi-dimensional market data fusion and real-time prediction technology.