Dynamic Knowledge Integration
Background: Large language models are increasingly used as semantic priors in time-series forecasting, yet the translation of numerical data into semantic prompts often relies on static, rule-based templates.
Question / Future Work: Determining effective methods for integrating dynamic, unstructured, and heterogeneous external knowledge (e.g., news, event logs) directly into forecasting frameworks, moving beyond the current reliance on rule-based prompting templates.
Why It Matters: This is a fundamental barrier to evolving LLM-based time series forecasting from static statistical assistance to dynamic, reasoning-driven adaptive systems.
Metadata & Links
- created_at
- 2026-04-17T05:45:21Z