Research map / Time series / Foundation models for time series
Language in the Flow of Time: Time-Series-Paired Texts Weaved into a Unified Temporal Narrative
Zihao Li, Xiao Lin, Zhining Liu, Jiaru Zou, Ziwei Wu, Lecheng Zheng, Dongqi Fu, Yada Zhu, Hendrik Hamann, Hanghang Tong, Jingrui He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add paired texts as auxiliary variables (TaTS) to iTransformer | iTransformer · Time-MMD Environment | MSE: 0.441 -> 0.267 | Improved |
| add paired texts as auxiliary variables (TaTS) to PatchTST | PatchTST · Time-MMD Economy | MSE: 0.017 -> 0.009 | Improved |
| use TaTS instead of MM-TSFLib text fusion | iTransformer · Time-MMD Environment | MSE: 0.421 -> 0.267 | Improved |
| add paired texts via TaTS | iTransformer · Time-MMD Energy | MAE: 0.375 -> 0.376 | No clear effect |
| add TaTS to imputation | PatchTST · Time-MMD Traffic imputation | MSE: 0.210 -> 0.131 | Improved |
| replace MLP projection with gated residual fusion (ablation) | TaTS · Time-MMD Climate | MSE: 0.992 -> 0.998 | No clear effect |
| replace MLP projection with cross-attention fusion (ablation) | TaTS · Time-MMD Climate | MSE: 0.992 -> 0.989 | No clear effect |
| keep texts aligned with timestamps instead of shuffling them (ablation) | iTransformer + TaTS · Time-MMD Climate | MSE: 1.242 -> 1.028 | Improved |
| keep texts aligned with timestamps instead of shuffling them (ablation) | iTransformer + TaTS · Time-MMD Traffic | MSE: 0.223 -> 0.187 | Improved |
| avoid randomly dropping 25% of texts (ablation) | iTransformer + TaTS · Time-MMD Traffic | MSE: 0.203 -> 0.187 | Improved |
| drop part of extremely noisy texts (20% corrupted instead of fully corrupted) (ablation) | iTransformer + TaTS · Time-MMD Climate | MSE: 1.242 -> 1.201 | Improved |
| use a larger text encoder (BERT 110M to GPT-2 to LLaMA2 7B) (ablation) | TaTS · Time-MMD datasets | MSE | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GPT4MTS — GPT4MTS: Prompt-based Large Language Model for Multimodal Time-series Forecasting (AAAI 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · similarity 0.91)
- TRACE: Grounding Time Series in Context for Multimodal Embedding and Retrieval (NeurIPS 2025 · similarity 0.90)
- VerbalTS: Generating Time Series from Texts (ICML 2025 · similarity 0.90)
- Empowering Time Series Analysis with Large-Scale Multimodal Pretraining (NeurIPS 2026 · similarity 0.89)
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · similarity 0.88)
- Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
