Research on multi-dimensional data mining and economic model construction for intelligent decision-making

Under the background of deep integration of digital economy and artificial intelligence, the traditional economic decision-making model is difficult to adapt to the new situation because of data explosion and increased uncertainty. Therefore, this paper puts forward an intelligent decision-making methodology integrating data, model and decision-making, aiming at solving three scientific problems: the difficulty of multi-dimensional data fusion, the lack of dynamic adaptation of economic model and the bottleneck of quantitative decision-making efficiency. In this paper, a cross-modal feature fusion method driven by knowledge graph (KG) is studied. Unstructured text entities are extracted by BERT-Transformer architecture, and the embedding space mapping between structured data and KG is realized by graph convolution network (GCN). At the same time, a double-layer coupling hybrid model architecture is designed, which combines the data-driven TCN-LSTM network with the core equation of the new Keynesian DSGE model, and introduces the gated attention module to realize dynamic coupling. In addition, a dynamic parameter calibration engine is built, and Bayesian particle filter combined with Gaussian mixture model with adaptive bandwidth is used to update the parameters in real time, which enhances the response ability of the system to emergencies. The experimental results show that the KG-Hybrid model proposed in this paper is significantly superior to the traditional model in risk early warning performance, dynamic adaptability and interpretability, which provides reliable technical support for intelligent decision-making systems.