Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Time-MQA: Time Series Multi-Task Question Answering with Context Enhancement

Created by
  • Haebom

Author

Yaxuan Kong, Yiyuan Yang, Yoontae Hwang, Wenjie Du, Stefan Zohren, Zhangyang Wang, Ming Jin, Qingsong Wen

Outline

This paper presents Time Series Multi-Task Question Answering (Time-MQA), an integrated framework for diverse temporal time series data (e.g., finance, healthcare, energy, etc.). Time-MQA supports natural language questions that enable open-ended question answering, including numerical analysis tasks and inference. At its core is a large-scale dataset, TSQA, containing approximately 200,000 question-answer pairs derived from diverse temporal time series data (e.g., environment, transportation, etc.). TSQA handles temporal time series of various lengths and facilitates the development of robust models. In addition, we demonstrate that large-scale language models, such as Mistral 7B, Llama-3 8B, and Qwen-2.5 7B, can be pre-trained on the TSQA dataset to enhance temporal time series inference capabilities and enable more sophisticated and intuitive interactions with temporal data beyond simple numerical tasks. The TSQA dataset, models, user study questionnaires for evaluation, and other related materials are open-sourced.

Takeaways, Limitations

•
Takeaways:
◦
We present Time-MQA, a novel framework that comprehensively handles various tasks (e.g., forecasting, anomaly detection, open-ended question answering) on time series data.
◦
Activating research on temporal time series analysis through the release of a large-scale temporal time series question-answering dataset, TSQA.
◦
A novel pre-training method is presented to improve the temporal time series inference capability of large-scale language models.
◦
Improving research accessibility through open source disclosure.
•
Limitations:
◦
Despite the diversity of the TSQA dataset, there may be biases toward specific domains or time series patterns.
◦
Despite the performance improvements of large-scale language models, there is a possibility that they are vulnerable to complex or ambiguous questions.
◦
Lack of detailed description of the scale and methodology of the user study.
👍
Made with Slashpage