Optimizing Economic Decision-Making Models through Multimodal Data Processing in Intelligent Systems

Under the background of the increasingly diversified global economic environment and the ever-increasing multi-source heterogeneous data flow, how to make efficient use of the intelligent system technology to synergize multi-modal data to provide the optimal selection scheme for economic decision-making model has become a pressing problem to solve. This suggests a complex model for the manner of economic decisionmaking. In the first stage, text and image data were extracted from features, while the continuous numerical data was taking advantage of time series model LSTM, dynamically tracking and predicting. Subsequently, a Transformer-based self-attention mechanism is introduced into the network architecture to mine the correlation between cross-modalities and cross-nodes to realize the deep integration of multi-modal information. Simultaneously, both the value network and the strategy network contribute to learning from varied of the economic environment to form an integrated closed loop from data collection, feature extraction to policy generation. In the stage of model training and optimization, it not only preserves the advantages of traditional statistical model in interpretability, but also makes use of the iterative interaction between deep neural network and reinforcement learning to effectively enhanced the generalization ability of the model and of the decision-making efficiency. The empirical findings indicate that, in economic situations, model outperforms standard model through enhanced decision accuracy and response speed.