Time series forecasting improved by using texts as extra data channels
TAC-Time: Texts as Channels For Multimodal Time Series Forecasting
Computation and LanguageArtificial IntelligenceMachine LearningMultimedia
Summary
Time series forecasting usually looks only at numbers changing over time, but this misses useful information from related texts. The authors created TAC-Time, a method that turns texts into time-based signals matching the number data so both can be analyzed together. This helps the model better understand patterns and time-related changes from texts, making forecasts more accurate and efficient. Their tests on real data show TAC-Time beats previous methods that used text less effectively.
What this means in practice
- •For financial analysts: Improve stock or economic forecasts by jointly modeling market texts and numerical indicators as temporal data channels.$Commercial implications: Enables development of forecasting tools that combine news texts with financial data for more accurate investment decisions.
- •For energy grid operators: Enhance demand forecasting by integrating textual reports with sensor readings as aligned time series signals.
Authors
Jiayi Liang, Xiaotian Gu, Xinyu Xie, Yuanbin Wu, Xiaoling Wang
Abstract
Most existing time series forecasting methods rely solely on numerical observations, overlooking rich contextual information from auxiliary texts. Recent multimodal approaches attempt to incorporate textual signals, but they often treat text as static features or use large language models as forecasting backbones, limiting their ability to capture temporal dynamics and increasing computational cost. To address these challenges, we propose TAC-Time, a unified framework that transforms textual information into additional temporal channels. By modeling text features jointly with numerical sequences in a shared temporal backbone, TAC-Time preserves temporal continuity and periodic structures while remaining efficient and scalable. This formulation also enables systematic interpretability analyses. We show strong cross-modal dependencies through attention and frequency-domain analyses, and identify predictive textual signals whose correlation-aware alignment yields partial forecasting improvements. Extensive experiments on real-world multimodal benchmarks demonstrate that TAC-Time outperforms prior methods.