TSRBENCH
收藏资源简介:
TSRBENCH是由多所顶尖高校联合构建的大规模多模态时间序列推理基准数据集,包含来自14个领域的4125个问题,涵盖感知、推理、预测和决策四大核心能力维度。数据集通过精心设计的15250条时间序列数据,支持文本、图像及混合模态输入,旨在全面评估通用模型在复杂时序场景下的多任务处理能力。其创建过程严格遵循文本-时序对齐和领域多样性原则,特别适用于金融、医疗、工业等关键领域的时间序列理解与决策支持研究。
TSRBENCH is a large-scale multimodal time series reasoning benchmark dataset jointly developed by several top-tier universities. It encompasses 4,125 questions spanning 14 distinct domains, covering four core capability dimensions: perception, reasoning, prediction, and decision-making. Equipped with 15,250 meticulously curated time series data samples, the dataset supports text, image, and hybrid modal inputs, and is designed to comprehensively evaluate the multi-task processing abilities of general-purpose models in complex temporal scenarios. Its development strictly adheres to the principles of text-temporal alignment and domain diversity, making it particularly well-suited for research on time series understanding and decision support in critical sectors such as finance, healthcare, and industry.

- 1TSRBench: A Comprehensive Multi-task Multi-modal Time Series Reasoning Benchmark for Generalist Models马里兰大学帕克分校; 伊利诺伊大学厄巴纳-香槟分校; 加州大学圣地亚哥分校; 穆罕默德·本·扎耶德人工智能大学 · 2026年



