ChronoSteer: Bridging Large Language Model and Time Series Foundation Model via Synthetic Data

TL;DR

ChronoSteer is a novel multimodal time series forecasting model that integrates large language models and time series foundation models via synthetic data, significantly improving prediction accuracy by leveraging textual information.

Contribution

This work introduces ChronoSteer, a framework that bridges LLMs and TSFMs using synthetic data and textual revision instructions, enabling multimodal forecasting with enhanced accuracy.

Findings

Achieves 25.7% improvement over unimodal models

Attains 22.5% higher accuracy than previous multimodal methods

Constructs a high-quality multimodal time series benchmark

Abstract

Conventional forecasting methods rely on unimodal time series data, limiting their ability to exploit rich textual information. Recently, large language models (LLMs) and time series foundation models (TSFMs) have demonstrated powerful capability in textual reasoning and temporal modeling, respectively. Integrating the strengths of both to construct a multimodal model that concurrently leverages both temporal and textual information for future inference has emerged as a critical research challenge. To address the scarcity of event-series paired data, we propose a decoupled framework: an LLM is employed to transform textual events into revision instructions, which are then used to steer the output of TSFM. To implement this framework, we introduce ChronoSteer, a multimodal TSFM that can be steered through textual revision instructions, effectively bridging LLM and TSFM. Moreover, to mitigate the shortage of cross-modal instruction-series paired data, we devise a two-stage training strategy based on synthetic data. In addition, we also construct a high-quality multimodal time series forecasting benchmark to address the information leakage concerns during evaluation. After integrating with an LLM, ChronoSteer, which is trained exclusively on synthetic data, achieves a 25.7% improvement in prediction accuracy compared to the unimodal backbone and a 22.5% gain over the previous state-of-the-art multimodal method.