Research on Commodity Demand Forecast Based on Big Data of Douyin E-commerce

In the era of accelerated digital transformation, e-commerce has become a cornerstone of modern economic systems. Leveraging the vast reserves of e-commerce big data for accurate demand forecasting is not only crucial for optimizing operational efficiency but also indispensable in today’s highly competitive markets. This study centers on the complex and large-scale data generated by Douyin e-commerce and develops a sophisticated demand forecasting model based on Bidirectional Long Short-Term Memory (LSTM) neural networks. Through detailed analysis of demand fluctuations, the model was rigorously trained and validated. Empirical results from simulation experiments are highly promising. In short-term (7-day) forecasting, some products achieved a prediction accuracy of 100%, with the overall Mean Absolute Error (MAE) at .17 and Root Mean Square Error (RMSE) at .47. For the more challenging 30-day prediction task, the model still demonstrated strong performance (MAE= .85, RMSE=8.23). These results highlight the model's robustness and generalizability across varying forecasting horizons. From a theoretical standpoint, this research enriches the literature on e-commerce demand prediction by introducing an advanced deep learning approach tailored to the dynamics of short-video-based online platforms. Practically, the model provides enterprises with a reliable tool to reduce inventory costs caused by over- or under-stocking, improve customer satisfaction through timely fulfillment, and enhance competitive advantage. Despite the inherent challenges posed by shifting user preferences, price volatility, and seasonal trends, the model maintains high predictive accuracy and adaptability. This study thus contributes meaningfully to both academic inquiry and the practical advancement of China’s e-commerce ecosystem.