TimeSage-MT
收藏资源简介:
TimeSage-MT是由多个研究机构联合创建的多轮对话基准数据集,旨在评估智能体在时间序列分析中的推理能力。该数据集包含240个任务和2,680个对话轮次,覆盖8个现实领域,数据来源于65个白名单真实世界时间序列源,通过可复现的生成流程构建。其创建过程涉及数据选择、分析路径规划、对话生成及多层质量控制,确保任务的可验证性和严谨性。该数据集主要应用于评估时间序列智能体系统,旨在解决现有基准在模拟实际分析工作流、多轮交互、跨能力诊断等方面的不足,推动可靠的时间序列决策支持系统发展。
TimeSage-MT is a multi-turn dialogue benchmark dataset jointly created by multiple research institutions, designed to evaluate the reasoning capabilities of AI Agents in time series analysis. This dataset contains 240 tasks and 2,680 dialogue turns, covering 8 real-world domains, with data sourced from 65 whitelisted real-world time series sources and constructed via a reproducible generation pipeline. Its creation process involves data selection, analysis path planning, dialogue generation and multi-layer quality control, ensuring the verifiability and rigor of the tasks. This dataset is mainly used to evaluate time series AI Agent systems, aiming to address the shortcomings of existing benchmarks in simulating actual analysis workflows, multi-turn interactions and cross-capability diagnosis, so as to promote the development of reliable time series decision support systems.
- 数据集名称:TimeSage-MT
- 许可协议:Apache-2.0
- 状态说明:该数据集标注为“即将到来”(Upcoming),暂未提供详细的数据内容、样本或使用说明。

- 1TimeSage-MT: A Multi-Turn Benchmark for Evaluating Agentic Time Series Reasoning牛津大学; 爱因霍芬理工大学; 格里菲斯大学; Squirrel Ai Learning; 华东师范大学; VulpiVox Intelligence · 2026年



