Sales Forecasting for Cultural and Creative Products Integrating Consumer Behavior and Sentiment Data: A Multimodal Optimization Strategy Based on Support Vector Machines

Abstract

With the rapid advancements in data technologies, sales forecasting is transitioning from traditional time-series models to multidimensional data-driven approaches [1]. Existing studies predominantly focus on single-modal data, which inadequately captures the deep associations between consumer behavior and sentiment data, particularly in the cultural and creative product domain characterized by high demand volatility and emotional dependency. This study proposes a multimodal sales forecasting model based on Support Vector Machines (SVM), integrating consumer behavior data (e.g., web search volume, repurchase rates) and sentiment data (e.g., sentiment scores, facial expressions) to enhance prediction accuracy and adaptability in dynamic market environments. By collecting and preprocessing behavior and sentiment data, constructing a multimodal input space, and applying SVM for nonlinear modeling and hyperparameter optimization, the experimental results demonstrate the proposed model's superiority over traditional methods (e.g., single-modal SVM and ARIMA). The model achieves lower prediction errors (MSE = 98.5, MAE = 7.3) and higher adaptability (R² = 0.89), with a forecasting deviation of only 6.2% during the "Singles' Day" promotional period, highlighting its potential application in dynamic markets. This study contributes by introducing an innovative multimodal integration approach, validating the role of sentiment data in sales forecasting for cultural and creative products, and showcasing the practical value of the model as theoretical support for real-time forecasting system development.