Time-MMD
收藏资源简介:
Time-MMD是由佐治亚理工学院创建的首个多域多模态时间序列数据集,涵盖了9个主要数据域。该数据集通过精心选择的数据源和严格的过滤步骤确保了细粒度的模态对齐,并从文本中分离事实与预测,确保所有截止日期更新至2024年5月。Time-MMD的创建过程涉及从多个分散源收集文本数据,并通过先进的语言模型进行预处理,以确保数据质量和精确对齐。该数据集旨在通过多模态扩展显著推进时间序列分析,特别是在需要结合文本和数值数据的领域,如流行病学和经济预测。
Time-MMD is the first multi-domain, multi-modal time series dataset developed by the Georgia Institute of Technology, covering nine major data domains. This dataset ensures fine-grained modal alignment via carefully curated data sources and rigorous filtering steps, distinguishes factual content from predictive statements in text, and guarantees that all deadlines are updated to May 2024. The creation process of Time-MMD involves collecting text data from multiple decentralized sources and preprocessing it with state-of-the-art language models to ensure data quality and precise alignment. This dataset aims to significantly advance time series analysis through multi-modal expansion, particularly in domains requiring the integration of textual and numerical data, such as epidemiology and economic forecasting.

- 1Time-MMD: A New Multi-Domain Multimodal Dataset for Time Series Analysis佐治亚理工学院 · 2024年



