Time-MQA: Time Series Multi-Task Question Answering with Context Enhancement
This addresses the problem of limited task diversity in time series analysis for domains like finance and healthcare, though it is incremental as it builds on existing large language models.
The paper tackles the narrow focus of existing time series methods by introducing Time-MQA, a unified framework for multi-task question answering, and the TSQA dataset with ~200k question-answer pairs, showing that continual pre-training on this dataset enhances time series reasoning in large language models.
Time series data are foundational in finance, healthcare, and energy domains. However, most existing methods and datasets remain focused on a narrow spectrum of tasks, such as forecasting or anomaly detection. To bridge this gap, we introduce Time Series Multi-Task Question Answering (Time-MQA), a unified framework that enables natural language queries across multiple time series tasks - numerical analytical tasks and open-ended question answering with reasoning. Central to Time-MQA is the TSQA dataset, a large-scale dataset containing $\sim$200k question-answer pairs derived from diverse time series spanning environment, traffic, etc. This comprehensive resource covers various time series lengths and promotes robust model development. We further demonstrate how continually pre-training large language models (Mistral 7B, Llama-3 8B, and Qwen-2.5 7B) on the TSQA dataset enhanced time series reasoning capabilities, moving beyond mere numeric tasks and enabling more advanced and intuitive interactions with temporal data. The complete TSQA dataset, models, user study questionnaires for evaluation, and other related materials have been open-sourced.