What If TSF (WIT)
收藏资源简介:
What If TSF (WIT)是由首尔国立大学团队构建的多模态时间序列预测基准数据集,旨在评估模型在专家构建的未来情景文本引导下的预测能力。该数据集涵盖政治、社会、能源和经济四大领域,包含5,352条样本,每条样本整合了历史时间序列数据、静态领域描述及动态未来情景文本。数据来源于权威新闻媒体和机构报告,经过LLM辅助预处理和专家验证,确保文本与时间序列对齐且去标识化。其核心应用为解决传统预测方法无法利用外部文本语境的问题,支持短期/长期预测及反事实推理任务,为多模态时序分析提供严谨测试平台。
What If TSF (WIT) is a multimodal time series forecasting benchmark dataset constructed by the team from Seoul National University, which aims to evaluate the predictive performance of models when guided by expert-constructed future scenario texts. This dataset covers four core domains: politics, society, energy and economics, and contains 5,352 samples. Each sample integrates historical time series data, static domain descriptions and dynamic future scenario texts. The data is sourced from authoritative news media and institutional reports, and has undergone LLM-assisted preprocessing and expert validation to ensure that the texts are aligned with the corresponding time series and de-identified. Its core application is to address the limitation that traditional forecasting methods cannot leverage external textual context, supporting short-term/long-term forecasting and counterfactual reasoning tasks, and providing a rigorous testbed for multimodal time series analysis.
WhatIfTSF 数据集概述
数据集名称
WhatIfTSF
数据集核心描述
WhatIfTSF 是一个基准数据集,旨在将时间序列预测任务重新定义为一种由场景引导的多模态预测问题。

- 1What If TSF: A Benchmark for Reframing Forecasting as Scenario-Guided Multimodal Forecasting首尔国立大学·数据科学研究生院 · 2026年



