Semantics-Enhanced Retrieval-Augmented Time Series Forecasting

In three linesSERAF, a time series forecasting framework, combines retrieval of historical segments with self-generated textual descriptions. Multimodal approach tested on 7 real-world datasets to improve predictions beyond numerical similarity alone.

Summary generated by Claude — human-verified