[Paper Review] EventCast: Hybrid Demand Forecasting in E-Commerce with LLM-Based Event Knowledge

Congcong Hu, Yuang Shi+5 more|arXiv (Cornell University)|Feb 7, 2026

Forecasting Techniques and Applications0 citations

TL;DR

This paper proposes a hybrid demand forecasting framework for e-commerce that utilizes large language model–based event knowledge to enhance forecasting.

ABSTRACT

Demand forecasting is a cornerstone of e-commerce operations, directly impacting inventory planning and fulfillment scheduling. However, existing forecasting systems often fail during high-impact periods such as flash sales, holiday campaigns, and sudden policy interventions, where demand patterns shift abruptly and unpredictably. In this paper, we introduce EventCast, a modular forecasting framework that integrates future event knowledge into time-series prediction. Unlike prior approaches that ignore future interventions or directly use large language models (LLMs) for numerical forecasting, EventCast leverages LLMs solely for event-driven reasoning. Unstructured business data, which covers campaigns, holiday schedules, and seller incentives, from existing operational databases, is processed by an LLM that converts it into interpretable textual summaries leveraging world knowledge for cultural nuances and novel event combinations. These summaries are fused with historical demand features within a dual-tower architecture, enabling accurate, explainable, and scalable forecasts. Deployed on real-world e-commerce scenarios spanning 4 countries of 160 regions over 10 months, EventCast achieves up to 86.9% and 97.7% improvement on MAE and MSE compared to the variant without event knowledge, and reduces MAE by up to 57.0% and MSE by 83.3% versus the best industrial baseline during event-driven periods. EventCast has deployed into real-world industrial pipelines since March 2025, offering a practical solution for improving operational decision-making in dynamic e-commerce environments.

Motivation & Objective

• Motivate the need for improved e-commerce demand forecasting.
• Propose a hybrid approach that integrates LLM-derived event knowledge with traditional forecasting methods.
• Demonstrate how event knowledge can be incorporated into forecasting workflows in e-commerce.
• Provide a pathway for leveraging LLMs to augment demand forecasting with event context.

Proposed method

• Introduce a hybrid forecasting framework combining traditional demand models with LLM-derived event knowledge.
• Extract and organize event-related knowledge from LLMs to inform forecasts.
• Integrate event signals into the forecasting pipeline to adjust demand predictions during events.
• Leverage sources and techniques aligned with information extraction and enterprise applications.

Experimental results

Research questions

• RQ1 How can event knowledge from large language models be represented for forecasting purposes?
• RQ2 Does integrating LLM-derived event knowledge improve e-commerce demand forecasts compared to traditional methods?
• RQ3 What is the best way to fuse event signals with existing forecasting models?
• RQ4 What are the practical considerations for deploying LLM-based event knowledge in e-commerce forecasting pipelines?

Key findings

• No quantitative findings are provided in the available excerpt.
• The work centers on presenting a hybrid approach leveraging LLM-based event knowledge for demand forecasting in e-commerce.
• The paper emphasizes the integration of event information into forecasting workflows.
• It situates the contribution at the intersection of e-commerce demand forecasting, large language models, and information extraction.