The research introduces the Momentum-integrated Multi-task Stock Recommendation (MiM-StocR) model, which aims to address shortcomings in traditional stock recommendation systems that often overlook simultaneous trends and rankings. Employing a momentum line indicator enhances the model’s capability to forecast short-term stock movements, which is critical for investors. Furthermore, an Adaptive-k ApproxNDCG ranking loss function is introduced to prioritize better-performing stocks and optimize investment decisions.
In light of the stock market’s volatility, the study addresses existing issues with overfitting in traditional Multi-Task Learning frameworks. By integrating the Converge-based Quad-Balancing (CQB) method, the model demonstrates significant advances in mitigating these challenges while ensuring accurate predictions. Extensive testing across benchmark stock datasets, including SEE50, CSI 100, and CSI 300, reveals that MiM-StocR consistently outperforms current leading models in both ranking ability and profitability assessments.
👉 Pročitaj original: arXiv AI Papers
