STReasoner: Empowering LLMs for Spatio-Temporal Reasoning in Time Series via Spatial-Aware Reinforcement Learning

TL;DR

This paper introduces STReasoner, a model that enhances large language models for explicit spatio-temporal reasoning in time series data, using a new benchmark and a spatial-aware reinforcement learning algorithm.

Contribution

The paper presents STReasoner, a novel approach integrating spatial and temporal reasoning in LLMs, along with ST-Bench, a new benchmark for evaluating such reasoning capabilities.

Findings

STReasoner achieves 17-135% accuracy improvements over baseline models.

The approach generalizes well to real-world spatio-temporal data.

S-GRPO reinforcement learning algorithm effectively promotes spatially grounded logic.

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 0.004X the cost of proprietary models and generalizes robustly to real-world data.