Multi-Modality Conditional Diffusion Model for Time Series Forecasting of Live Sales Volume

Accurately predicting the volume of sales in live broadcast is crucial for e-commerce. Despite the success of the current sales volume prediction models, theirs application is significantly constrained by the absence of high-quality live sales volume data. In this paper, we introduce a novel application of the diffusion model in live broadcast sales forecasting, leveraging multi-modal information as a generative condition to enhance prediction quality. We transform historical live broadcasts into a sequence of image and text data, subsequently embedded via ResNet and Bert respectively, in which allowing us to generate live sales volume series for downstream tasks. To enhance the data quality, we resort to currently popular diffusion models. To our knowledge, this is the first application of time-series diffusion generation to live streaming data, enriching our comprehension of diverse video and text data in live broadcast settings. We conduct extensive experiments including generate effect validation, modal elimination and visualization, the prominent results demonstrates the effectiveness of our proposal.