TY - GEN
T1 - LoFT-LLM
T2 - 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1, KDD 2026
AU - You, Jiacheng
AU - Yang, Jingcheng
AU - Xie, Yuhang
AU - Wu, Zhongxuan
AU - Li, Xiucheng
AU - Li, Feng
AU - Wang, Pengjie
AU - Xu, Jian
AU - Zheng, Bo
AU - Chen, Xinyang
N1 - Publisher Copyright:
© 2026 Owner/Author.
PY - 2026/4/20
Y1 - 2026/4/20
N2 - Time-series forecasting in real-world applications such as finance and energy often faces challenges due to limited training data and complex, noisy temporal dynamics. Existing deep forecasting models typically supervise predictions using full-length temporal windows, which include substantial high-frequency noise and obscure long-term trends. Moreover, auxiliary variables containing rich domain-specific information are often underutilized, especially in few-shot settings. To address these challenges, we propose LoFT-LLM, a frequency-aware forecasting pipeline that integrates low-frequency learning with semantic calibration via a large language model (LLM). Firstly, a Patch Low-Frequency forecasting Module (PLFM) extracts stable low-frequency trends from localized spectral patches. Secondly, a residual learner then models high-frequency variations. Finally, a fine-tuned LLM refines the predictions by incorporating auxiliary context and domain knowledge through structured natural language prompts. Extensive experiments on financial and energy datasets demonstrate that LoFT-LLM significantly outperforms strong baselines under both full-data and few-shot regimes, delivering superior accuracy, robustness, and interpretability.
AB - Time-series forecasting in real-world applications such as finance and energy often faces challenges due to limited training data and complex, noisy temporal dynamics. Existing deep forecasting models typically supervise predictions using full-length temporal windows, which include substantial high-frequency noise and obscure long-term trends. Moreover, auxiliary variables containing rich domain-specific information are often underutilized, especially in few-shot settings. To address these challenges, we propose LoFT-LLM, a frequency-aware forecasting pipeline that integrates low-frequency learning with semantic calibration via a large language model (LLM). Firstly, a Patch Low-Frequency forecasting Module (PLFM) extracts stable low-frequency trends from localized spectral patches. Secondly, a residual learner then models high-frequency variations. Finally, a fine-tuned LLM refines the predictions by incorporating auxiliary context and domain knowledge through structured natural language prompts. Extensive experiments on financial and energy datasets demonstrate that LoFT-LLM significantly outperforms strong baselines under both full-data and few-shot regimes, delivering superior accuracy, robustness, and interpretability.
KW - few shot
KW - large language models
KW - low-frequency learning
KW - time series forecasting
UR - https://www.scopus.com/pages/publications/105038078974
U2 - 10.1145/3770854.3780245
DO - 10.1145/3770854.3780245
M3 - 会议稿件
AN - SCOPUS:105038078974
T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
SP - 1809
EP - 1820
BT - KDD 2026 - Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1
PB - Association for Computing Machinery
Y2 - 9 August 2026 through 13 August 2026
ER -