LLaTiSA: Towards Difficulty-Stratified Time Series Reasoning from Visual Perception to Semantics
For researchers in time series analysis and LLMs, this work provides a unified framework and strong baseline for time series reasoning, addressing fragmented benchmarks and evaluation ambiguities.
The paper formalizes Time Series Reasoning (TSR) with a four-level taxonomy and introduces HiTSR, a dataset of 83k samples with Chain-of-Thought trajectories. The proposed model LLaTiSA integrates visual patterns and numerical tables, achieving superior performance and robust out-of-distribution generalization across diverse TSR tasks.
Comprehensive understanding of time series remains a significant challenge for Large Language Models (LLMs). Current research is hindered by fragmented task definitions and benchmarks with inherent ambiguities, precluding rigorous evaluation and the development of unified Time Series Reasoning Models(TSRMs). To bridge this gap, we formalize Time Series Reasoning (TSR) via a four-level taxonomy of increasing cognitive complexity. We introduce HiTSR, a hierarchical time series reasoning dataset comprising 83k samples with diverse task combinations and verified Chain-of-Thought (CoT) trajectories. Leveraging HiTSR, we propose LLaTiSA, a strong TSRM that integrates visualized patterns with precision-calibrated numerical tables to enhance the temporal perception of Vision-Language Models (VLMs). Through a multi-stage curriculum fine-tuning strategy, LLaTiSA achieves superior performance and exhibits robust out-of-distribution generalization across diverse TSR tasks and real-world scenarios. Our code is available at https://github.com/RainingNovember/LLaTiSA.