This is the authors' abstract. We don't add key points for this paper.
Stock price movement prediction is pivotal yet inherently challenging in financial markets, mainly due to difficulties in effectively integrating multimodal data and modeling intricate, dynamic inter-stock relationships. This paper proposes FFIR-Net, an innovative deep learning framework for precise prediction with two core functions: Feature Fusion and Inter-stock Relationship Learning. It has three key components: Tensor Fusion that fuses daily trading data and market sentiment information through attention-based alignment, Multiscale Feature Mixer that captures temporal evolution patterns by decomposing historical data into multiscale subsequences, and HGAT-GE that models intricate inter-stock relationships. Its three major contributions include an Inter-Stock Relationship Learning module revealing hidden multiscale dependencies via HGAT-GE and dynamic hypergraph generation, an end-to-end framework without manual feature engineering, and validation on CSI100E and S&P500 datasets. FFIR-Net outperforms state-of-the-art (SOTA) methods in both classification metrics and practical trading profitability, and displays robust adaptability to bull, bear, and sideways markets.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: