UniBizNet: A Unified Business Intelligence Network for Strategic Multimodal Data Integration and Predictive Analysis

This paper presents UniBizNet, a Unified Business Intelligence Network for strategic multimodal data integration and predictive analysis. UniBizNet is designed to effectively combine diverse data modalities, including customer sales, financial metrics, and customer sentiment, to provide robust predictions for business decision-making. The proposed architecture integrates Long Short-Term Memory blocks for sequential feature extraction, Transformer blocks for textual analysis, and Attention-Gated Units to dynamically weigh the importance of each feature, forming a comprehensive multimodal representation. Our methodology was evaluated on three tasks: sales forecasting, stock price movement prediction, and sentiment impact analysis on sales performance. For sales forecasting, UniBizNet achieved a Mean Absolute Error of 95.21, a Root Mean Squared Error of 135.45, and a Mean Absolute Percentage Error of 12.4%, outperforming state-of-the-art models, including ARIMA, LSTM, and Transformer, by reducing MAPE by at least 1.4%. For stock price movement prediction, the proposed model demonstrated superior performance with an MAE of 24.87, an RMSE of 34.12, and a directional accuracy of 83.1%, which is an improvement of 2.9% over the best-performing baseline model. In sentiment impact analysis, UniBizNet achieved a correlation coefficient of 0.78 and an F1 Score of 0.88, surpassing BERT and other baselines by at least 4.0%.