STReasoner: Empowering LLMs for Spatio-Temporal Reasoning in Time Series via Spatial-Aware Reinforcement Learning
STReasoner uses spatial-aware reinforcement learning to empower LLMs for spatio-temporal reasoning in time series, with large accuracy gains over proprietary models at low cost.
Published 20261 citationPaper ↗
Only vote on papers you've read. Sign in with GitHub to vote.
STReasoner delivers striking cost-efficient reasoning gains on the ST-Bench benchmark via spatially grounded reinforcement learning, though its synthetic multi-agent pipeline, wide accuracy spread, and unverified real-world generalization leave the source of those gains unclear.
Abstract
Spatio-temporal reasoning in time series involves the explicit synthesis of temporal dynamics, spatial dependencies, and textual context.This capability is vital for high-stakes decision-making in systems such as traffic networks, power grids, and disease propagation.However, the field remains underdeveloped because most existing works prioritize predictive accuracy over reasoning.To address the gap, we introduce ST-Bench, a benchmark consisting of four core tasks, including etiological reasoning, entity identification, correlation reasoning, and in-context forecasting, developed via a network SDE-based multi-agent data synthesis pipeline.We then propose STReasoner, which empowers LLM to integrate time series, graph structure, and text for explicit reasoning.To promote spatially grounded logic, we introduce S-GRPO, a reinforcement learning algorithm that rewards performance gains specifically attributable to spatial information.Experiments show that STReasoner achieves average accuracy gains between 17% and 135% at only 0.004× the cost of proprietary models and generalizes robustly to real-world data.Our code is available at https://github.com/LingFengGold/STReasoner.