Intelligent Portfolio Recommendation System Based on Multimodal Financial Data Fusion

Abstract

With the increasing diversity and complexity of financial data, how to effectively fuse multimodal information to improve the accuracy and robustness of investment decisions has become a research hotspot. This paper proposes an intelligent portfolio recommendation
system based on multimodal data fusion, introducing deep learning and attention mechanisms to achieve adaptive fusion and modeling of
multi-source data such as text sentiment and numerical time-series features. The system constructs a unified feature space through cross-modal
alignment technology and combines dynamic asset allocation algorithms with multi-objective optimization strategies to achieve an effective
balance between risk and return. Experimental results show that the system significantly outperforms traditional single-data models in a backtesting environment, especially in key indicators such as the Sharpe ratio and maximum drawdown, verifying the potential of multimodal fusion in improving portfolio performance.