Language in the Flow of Time: Time-Series-Paired Texts Weaved into a Unified Temporal Narrative
Organizations: University of Illinois Urbana-Champaign · Meta · IBM Research
Abstract
While many advances in time series models focus exclusively on numerical data, research on multimodal time series, particularly those involving contextual textual information, remains in its infancy. With recent progress in large language models and time series learning, we revisit the integration of paired texts with time series through the Platonic Representation Hypothesis, which posits that representations of different modalities converge to shared spaces. In this context, we identify that time-series-paired texts may naturally exhibit periodic properties that closely mirror those of the original time series. Building on this insight, we propose a novel framework, Texts as Time Series (TaTS), which considers the time-series-paired texts to be auxiliary variables of the time series. TaTS can be plugged into any existing numerical-only time series models and effectively enable them to handle time series data with paired texts. Through extensive experiments on both multimodal time series forecasting and imputation tasks across benchmark datasets with various existing time series models, we demonstrate that TaTS can enhance multimodal predictive performance without modifying model architectures. Our Code is available at https://github.com/iDEA-iSAIL-Lab-UIUC/TaTS.
Figures & tables
| Dataset | Monthly Sampled | Weekly Sampled | Daily Sampled | ||||||
| Agriculture | Climate | Economy | Security | Social Good | Traffic | Energy | Health | Environment | |
| Original | 0.026 | 0.025 | 0.022 | 0.049 | 0.027 | 0.035 | 0.307 | 0.233 | 0.302 |
| TS shuffled | 0.088 | 0.032 | 0.098 | 0.054 | 0.069 | 0.102 | 0.320 | 0.268 | 0.358 |
| Text shuffled | 0.106 | 0.037 | 0.099 | 0.053 | 0.072 | 0.104 | 0.312 | 0.277 | 0.364 |
| Models | iTransformer | PatchTST | Crossformer | DLinear | FEDformer | FiLM | Autoformer | Informer | Transformer | ||||||||||
| ( 2024c ) | ( 2023 ) | ( 2023 ) | ( 2023 ) | ( 2022b ) | ( 2022a ) | ( 2021 ) | ( 2021 ) | ( 2017 ) | |||||||||||
| Metric | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | |
| Agriculture | Uni-modal | 0.122 | 0.251 | 0.120 | 0.247 | 0.323 | 0.406 | 0.223 | 0.354 | 0.138 | 0.286 | 0.139 | 0.256 | 0.158 | 0.297 | 0.599 | 0.630 | 0.354 | 0.434 |
| MM-TSFLib | 0.112 | 0.230 | 0.114 | 0.233 | 0.218 | 0.313 | 0.218 | 0.355 | 0.131 | 0.275 | 0.140 | 0.258 | 0.158 | 0.288 | 0.313 | 0.414 | 0.249 | 0.352 | |
| TaTS (ours) | 0.109 | 0.229 | 0.114 | 0.235 | 0.212 | 0.312 | 0.214 | 0.351 | 0.131 | 0.276 | 0.135 | 0.251 | 0.125 | 0.266 | 0.255 | 0.348 | 0.191 | 0.313 | |
| Promotion | 2.7% | 0.4% | 0.0% | -0.9% | 2.8% | 0.3% | 1.8% | 0.8% | 0.0% | -0.4% | 2.9% | 2.0% | 20.9% | 7.6% | 18.5% | 15.9% | 23.3% | 11.1% | |
| Models | PatchTST | DLinear | FiLM | ||||
| ( 2023 ) | ( 2023 ) | ( 2022a ) | |||||
| Metric | MSE | MAE | MSE | MAE | MSE | MAE | |
| Climate | Uni-modal | 1.111 | 0.846 | 0.969 | 0.801 | 1.123 | 0.829 |
| MM-TSFLib | 1.010 | 0.821 | 0.963 | 0.802 | 1.130 | 0.833 | |
| TaTS (ours) | 0.878 | 0.720 | 0.912 | 0.757 | 0.820 | 0.718 | |
| Promotion | 13.1% | 12.3% | 5.3% | 5.5% | 27.0% | 13.4% | |
| Methods | TaTS (ours) | TaTS (ours) | TaTS (ours) | N-BEATS | N-HiTS | TCN | ChatTime | GPT4MTS | |||||||||
| + iTransformer | + PatchTST | + FiLM | ( 2020 ) | ( 2022 ) | ( 2018 ) | ( 2025 ) | ( 2024 ) | ||||||||||
| Datasets | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | |
| Time-MMD: Multimodal Time Series ( 2024a ) | Agriculture | 0.109 | 0.229 | 0.114 | 0.235 | 0.135 | 0.251 | 3.267 | 1.458 | 1.852 | 1.032 | 4.168 | 1.797 | 0.508 | 0.447 | 0.327 | 0.393 |
| Climate | 1.028 | 0.804 | 1.004 | 0.798 | 0.945 | 0.772 | 1.093 | 0.861 | 1.103 | 0.858 | 1.098 | 0.866 | 1.568 | 1.019 | 1.127 | 0.873 | |
| Economy | 0.008 | 0.077 | 0.009 | 0.079 | 0.009 | 0.080 | 1.010 | 0.920 | 0.444 | 0.584 | 5.546 | 2.349 | 0.049 | 0.166 | 0.014 | 0.096 | |
| Energy | 0.265 | 0.376 | 0.258 | 0.371 | 0.271 | 0.379 | 0.329 | 0.424 | 0.372 | 0.463 | 0.430 | 0.512 | 0.305 | 0.417 | 0.269 | 0.378 | |
| Dataset | Monthly Sampled | Weekly Sampled | Daily Sampled | ||||||
| Agriculture | Climate | Economy | Security | Social Good | Traffic | Energy | Health | Environment | |
| Original-Shuffled TT Ratio (%) | 26.8 | 72.4 | 22.3 | 91.5 | 38.2 | 34.6 | 97.1 | 85.6 | 83.6 |
| TaTS Improvement (%) | 20.70 | 18.03 | 64.80 | 4.05 | 10.99 | 16.78 | 3.62 | 19.51 | 36.00 |
| Dataset | Climate | Security | Traffic | # Parameters | ||||
| TaTS Settings | MSE | MAE | MSE | MAE | MSE | MAE | Fuser | Total |
| MLP projection (original) | 0.992 | 0.791 | 109.865 | 4.968 | 0.178 | 0.224 | 74988 | 6396658 |
| gated residual | 0.998 | 0.794 | 110.254 | 4.965 | 0.177 | 0.227 | 9396 | 6331066 |
| cross-attention | 0.989 | 0.792 | 109.729 | 4.976 | 0.178 | 0.221 | 99340 | 6421010 |
| Dataset | Climate | Security | Traffic | |||
| Settings | MSE | MAE | MSE | MAE | MSE | MAE |
| iTransformer + TaTS + original data | 1.028 | 0.804 | 112.05 | 5.151 | 0.187 | 0.217 |
| iTransformer + TaTS + corrupted data | 1.242 | 0.895 | 117.82 | 5.767 | 0.223 | 0.265 |
| iTransformer + uni-modal + original data | 1.183 | 0.871 | 115.94 | 5.660 | 0.213 | 0.238 |
| Dataset | Climate | Security | Traffic | |||
| Settings | MSE | MAE | MSE | MAE | MSE | MAE |
| TaTS + original data | 1.028 | 0.804 | 112.05 | 5.151 | 0.187 | 0.217 |
| TaTS + 10% text randomly dropped | 1.032 | 0.809 | 112.94 | 5.372 | 0.194 | 0.239 |
| TaTS + 25% text randomly dropped | 1.056 | 0.818 | 114.59 | 5.231 | 0.203 | 0.254 |
| Dataset | Climate | Security | Traffic | |||
| Settings | MSE | MAE | MSE | MAE | MSE | MAE |
| TaTS + corrupted data | 1.242 | 0.895 | 117.82 | 5.767 | 0.223 | 0.265 |
| TaTS + 60% corrupted data | 1.218 | 0.879 | 117.43 | 5.730 | 0.215 | 0.260 |
| TaTS + 20% corrupted data | 1.201 | 0.878 | 116.68 | 5.713 | 0.216 | 0.255 |
| uni-modal + original data | 1.183 | 0.871 | 115.94 | 5.660 | 0.213 | 0.238 |
Appendix figures & tables32 assets
Supplementary material from the paper’s appendix.
Appendix
| Dataset | Monthly Sampled | Weekly Sampled | Daily Sampled | ||||||
| Agriculture | Climate | Economy | Security | Social Good | Traffic | Energy | Health | Environment | |
| Default | 0.026 | 0.025 | 0.022 | 0.049 | 0.027 | 0.035 | 0.307 | 0.233 | 0.302 |
| Hann Window | 0.031 | 0.029 | 0.016 | 0.041 | 0.040 | 0.057 | 0.304 | 0.243 | 0.306 |
| Hamming Window | 0.054 | 0.025 | 0.016 | 0.041 | 0.037 | 0.046 | 0.305 | 0.242 | 0.305 |
| Blackman Window | 0.061 | 0.029 | 0.020 | 0.036 | 0.043 | 0.055 | 0.297 | 0.245 | 0.298 |
| Minmax Normalization | 0.032 | 0.035 | 0.025 | 0.043 | 0.031 | 0.049 | 0.345 | 0.291 | 0.332 |
| Dataset | Monthly Sampled | Weekly Sampled | Daily Sampled | ||||||
| Agriculture | Climate | Economy | Security | Social Good | Traffic | Energy | Health | Environment | |
| Default | 0.06 | 0.04 | 0.03 | 0.03 | 0.04 | 0.06 | 0.21 | 0.16 | 0.12 |
| Methods | TaTS (Ours) | TFHTS | MCD-TSF | TeR-TSF | |||||
| + iTransformer | ( 2025 ) | ( 2025b ) | ( 2025a ) | ||||||
| Datasets | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | |
| Time-MMD: Multimodal Time Series ( 2024a ) | Agriculture | 0.201 | 0.321 | 0.571 | 0.564 | 0.222 | 0.322 | 0.338 | 0.402 |
| Climate | 1.227 | 0.890 | 1.782 | 0.948 | 1.583 | 0.971 | 1.348 | 0.874 | |
| Energy | 0.214 | 0.321 | 0.290 | 0.403 | 0.153 | 0.293 | 0.202 | 0.324 | |
| Environment | 0.267 | 0.374 | 0.262 | 0.368 | 0.275 | 0.379 | 0.251 | 0.346 | |
| Monthly Sampled | Weekly Sampled | Daily Sampled | |||||||
| Agriculture | Climate | Economy | Security | Social Good | Traffic | Energy | Health | Environment | |
| BERT | 0.12 | 0.21 | 0.23 | 0.23 | 1.7 | 0.21 | 0.26 | 0.46 | 0.29 |
| GPT-2 | 0.18 | 0.25 | 0.31 | 0.28 | 2.5 | 0.30 | 0.28 | 0.55 | 0.35 |
| LLaMA-2-7B | 0.28 | 0.29 | 0.37 | 0.35 | 10.5 | 0.40 | 0.32 | 0.62 | 0.42 |
| Dataset Name/Domain | Prediction Length | Dimension | Frequency | Number of Samples | Timespan |
| Agriculture | {6, 8, 10, 12} | 1 | Monthly | 1983 - Present | |
| Climate | {6, 8, 10, 12} | 5 | Monthly | 1983 - Present | |
| Economy | {6, 8, 10, 12} | 3 | Monthly | 1989 - Present | |
| Energy | {12, 24, 36, 48} | 9 | Weekly | 1996 - Present | |
| Environment | {48, 96, 192, 336} | 4 | Daily | 1982 - 2023 | |
| Health | {12, 24, 36, 48} | 11 | Weekly | 1997 - Present |
| Company/Stock Name | Prediction Length | Dimension | Frequency | Number of Samples | Timespan |
| Delta Airlines (DAL) | {6, 8, 10, 12} | 1 | Bi-daily | 2009 - 2020 | |
| IBM (IBM) | {6, 8, 10, 12} | 1 | Bi-daily | 2016 - 2020 | |
| JPMorgan Chase (JPM) | {6, 8, 10, 12} | 1 | Bi-daily | 2018 - 2020 | |
| NVIDIA (NVDA) | {6, 8, 10, 12} | 1 | Bi-daily | 2011 - 2020 | |
| Pfizer (PFE) | {6, 8, 10, 12} | 1 | Bi-daily | 2016 - 2020 | |
| Tesla (TSLA) | {6, 8, 10, 12} | 1 | Bi-daily | 2019 - 2020 |
| Dataset Name/Domain | Prediction Length | Dimension | Frequency | Number of Samples | Timespan |
| Bitcoin Price | {6, 8, 10, 12} | 1 | Daily | 2018 - 2021 | |
| Web Traffic | {6, 8, 10, 12} | 1 | Daily | 2015 - 2016 | |
| Electricity Demand | {6, 8, 10, 12} | 1 | Daily | 2019 - 2021 |
| Hyperparameter | Description | Value or Choices |
| batch_size | The batch size for training | 32 |
| criterion | The criterion for calculating loss | Mean Square Error (MSE) |
| learning_rate | The learning rate for the optimizer | {0.0001, 0.00005, 0.00001} |
| seq_len | Input sequence length | 24 |
| label_len | Start token length for prediction | 12 |
| prior_weight | Weight for prior combination | {0, 0.1, 0.2, 0.3, 0.5} |
| Method | Economy | Social Good | Traffic | |||
| MSE ( ) | MAE ( ) | MSE ( ) | MAE ( ) | MSE ( ) | MAE ( ) | |
| Uniformly Random | 5.673 | 2.356 | 2.059 | 1.230 | 1.207 | 0.995 |
| Uniformly Random | 11.535 | 2.879 | 8.860 | 2.404 | 3.794 | 1.618 |
| Normally Random | 9.284 | 2.878 | 2.926 | 1.374 | 3.163 | 1.511 |
| Exponentially Random | 4.521 | 1.960 | 3.724 | 1.564 | 1.141 | 0.911 |
| Using 1D Text only | 1.995 | 1.404 | 1.315 | 0.853 | 0.714 | 0.797 |
| Models | iTransformer | PatchTST | Crossformer | ||||||||||||||
| ( 2024c ) | ( 2023 ) | ( 2023 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Agriculture | Uni-modal | 6 | 0.077 | 0.200 | 0.274 | 0.090 | 0.014 | 0.074 | 0.197 | 0.269 | 0.090 | 0.014 | 0.222 | 0.331 | 0.412 | 0.136 | 0.031 |
| 8 | 0.104 | 0.228 | 0.315 | 0.100 | 0.017 | 0.104 | 0.234 | 0.317 | 0.104 | 0.018 | 0.304 | 0.406 | 0.496 | 0.168 | 0.044 | ||
| 10 | 0.142 | 0.273 | 0.372 | 0.119 | 0.024 | 0.136 | 0.262 | 0.357 | 0.112 | 0.021 | 0.357 | 0.435 | 0.530 | 0.176 | 0.049 | ||
| 12 | 0.167 | 0.301 | 0.400 | 0.128 | 0.026 | 0.168 | 0.294 | 0.396 | 0.124 | 0.025 | 0.409 | 0.451 | 0.582 | 0.181 | 0.054 | ||
| Models | iTransformer | PatchTST | Crossformer | ||||||||||||||
| ( 2024c ) | ( 2023 ) | ( 2023 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Energy | Uni-modal | 12 | 0.112 | 0.231 | 0.305 | 0.940 | 10.877 | 0.105 | 0.229 | 0.301 | 1.120 | 40.016 | 0.138 | 0.262 | 0.338 | 1.121 | 31.808 |
| 24 | 0.222 | 0.352 | 0.438 | 1.626 | 35.711 | 0.241 | 0.363 | 0.458 | 1.507 | 41.679 | 0.281 | 0.399 | 0.482 | 1.619 | 46.008 | ||
| 36 | 0.306 | 0.409 | 0.511 | 1.767 | 58.611 | 0.304 | 0.408 | 0.512 | 1.782 | 60.280 | 0.331 | 0.443 | 0.541 | 3.296 | 229.673 | ||
| 48 | 0.435 | 0.509 | 0.617 | 2.454 | 95.250 | 0.427 | 0.502 | 0.610 | 2.405 | 91.255 | 0.422 | 0.521 | 0.626 | 4.574 | 347.533 | ||
| Models | iTransformer | PatchTST | Crossformer | ||||||||||||||
| ( 2024c ) | ( 2023 ) | ( 2023 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Security | Uni-modal | 6 | 113.573 | 5.698 | 10.657 | 6.008 | 1020.167 | 108.795 | 5.119 | 10.430 | 4.255 | 549.239 | 124.972 | 6.077 | 11.179 | 2.435 | 113.813 |
| 8 | 115.878 | 5.743 | 10.765 | 5.977 | 1035.504 | 113.534 | 5.524 | 10.655 | 3.681 | 424.901 | 126.513 | 6.237 | 11.248 | 1.909 | 67.155 | ||
| 10 | 116.950 | 5.615 | 10.814 | 3.109 | 332.477 | 114.820 | 5.484 | 10.715 | 2.523 | 185.108 | 127.701 | 6.346 | 11.300 | 1.678 | 70.642 | ||
| 12 | 117.396 | 5.585 | 10.835 | 2.645 | 262.692 | 114.253 | 5.357 | 10.689 | 1.951 | 117.032 | 128.686 | 6.449 | 11.344 | 1.484 | 35.512 | ||
| Models | DLinear | FEDformer | FiLM | ||||||||||||||
| ( 2023 ) | ( 2022b ) | ( 2022a ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Agriculture | Uni-modal | 6 | 0.170 | 0.312 | 0.395 | 0.135 | 0.028 | 0.091 | 0.239 | 0.301 | 0.110 | 0.019 | 0.088 | 0.207 | 0.290 | 0.092 | 0.015 |
| 8 | 0.195 | 0.340 | 0.423 | 0.145 | 0.031 | 0.127 | 0.281 | 0.355 | 0.127 | 0.024 | 0.115 | 0.230 | 0.318 | 0.097 | 0.017 | ||
| 10 | 0.218 | 0.355 | 0.448 | 0.150 | 0.034 | 0.154 | 0.313 | 0.392 | 0.142 | 0.030 | 0.151 | 0.255 | 0.351 | 0.103 | 0.020 | ||
| 12 | 0.308 | 0.407 | 0.514 | 0.165 | 0.041 | 0.180 | 0.313 | 0.415 | 0.134 | 0.028 | 0.200 | 0.333 | 0.433 | 0.140 | 0.030 | ||
| Models | DLinear | FEDformer | FiLM | ||||||||||||||
| ( 2023 ) | ( 2022b ) | ( 2022a ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Energy | Uni-modal | 12 | 0.136 | 0.264 | 0.335 | 0.947 | 10.373 | 0.095 | 0.212 | 0.281 | 0.895 | 9.380 | 0.118 | 0.245 | 0.314 | 1.208 | 25.502 |
| 24 | 0.261 | 0.385 | 0.467 | 1.453 | 33.690 | 0.170 | 0.303 | 0.385 | 1.573 | 47.676 | 0.221 | 0.347 | 0.433 | 1.414 | 29.039 | ||
| 36 | 0.335 | 0.437 | 0.528 | 1.780 | 56.178 | 0.249 | 0.373 | 0.475 | 2.165 | 133.305 | 0.335 | 0.434 | 0.533 | 1.852 | 60.033 | ||
| 48 | 0.431 | 0.498 | 0.603 | 2.294 | 83.886 | 0.445 | 0.514 | 0.643 | 4.141 | 340.071 | 0.437 | 0.512 | 0.620 | 2.414 | 89.205 | ||
| Models | DLinear | FEDformer | FiLM | ||||||||||||||
| ( 2023 ) | ( 2022b ) | ( 2022a ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Security | Uni-modal | 6 | 107.046 | 4.531 | 10.346 | 3.489 | 324.650 | 112.106 | 4.830 | 10.588 | 4.456 | 571.690 | 115.130 | 5.587 | 10.730 | 5.029 | 637.394 |
| 8 | 107.700 | 4.649 | 10.378 | 2.992 | 295.719 | 113.897 | 5.200 | 10.672 | 3.223 | 355.929 | 112.097 | 5.034 | 10.588 | 3.228 | 336.503 | ||
| 10 | 109.360 | 4.759 | 10.458 | 2.449 | 216.866 | 117.484 | 5.467 | 10.839 | 2.563 | 216.459 | 112.267 | 4.901 | 10.596 | 2.310 | 188.201 | ||
| 12 | 112.343 | 4.904 | 10.599 | 1.980 | 149.507 | 114.452 | 5.135 | 10.698 | 2.697 | 329.415 | 122.711 | 6.425 | 11.077 | 3.554 | 437.883 | ||
| Models | Autoformer | Informer | Transformer | ||||||||||||||
| ( 2021 ) | ( 2021 ) | ( 2017 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Agriculture | Uni-modal | 6 | 0.109 | 0.255 | 0.329 | 0.116 | 0.022 | 0.451 | 0.557 | 0.626 | 0.240 | 0.071 | 0.229 | 0.326 | 0.420 | 0.133 | 0.032 |
| 8 | 0.135 | 0.278 | 0.366 | 0.126 | 0.026 | 0.569 | 0.633 | 0.702 | 0.270 | 0.087 | 0.328 | 0.432 | 0.502 | 0.177 | 0.045 | ||
| 10 | 0.173 | 0.311 | 0.413 | 0.138 | 0.031 | 0.618 | 0.633 | 0.724 | 0.263 | 0.088 | 0.358 | 0.422 | 0.510 | 0.166 | 0.045 | ||
| 12 | 0.213 | 0.345 | 0.457 | 0.149 | 0.036 | 0.756 | 0.698 | 0.789 | 0.284 | 0.101 | 0.500 | 0.555 | 0.633 | 0.224 | 0.064 | ||
| Models | Autoformer | Informer | Transformer | ||||||||||||||
| ( 2021 ) | ( 2021 ) | ( 2017 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Energy | Uni-modal | 12 | 0.161 | 0.282 | 0.362 | 1.403 | 24.893 | 0.165 | 0.289 | 0.373 | 1.491 | 35.971 | 0.124 | 0.247 | 0.320 | 1.021 | 8.908 |
| 24 | 0.279 | 0.416 | 0.509 | 2.632 | 158.931 | 0.248 | 0.380 | 0.475 | 2.916 | 185.671 | 0.291 | 0.403 | 0.494 | 1.951 | 101.566 | ||
| 36 | 0.364 | 0.464 | 0.564 | 2.384 | 93.988 | 0.368 | 0.492 | 0.585 | 4.022 | 356.146 | 0.329 | 0.435 | 0.542 | 3.964 | 428.707 | ||
| 48 | 0.471 | 0.550 | 0.664 | 3.106 | 149.374 | 0.457 | 0.539 | 0.640 | 5.613 | 589.886 | 0.445 | 0.537 | 0.642 | 3.824 | 285.963 | ||
| Models | Autoformer | Informer | Transformer | ||||||||||||||
| ( 2021 ) | ( 2021 ) | ( 2017 ) | |||||||||||||||
| Method | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | ||
| Security | Uni-modal | 6 | 115.544 | 5.268 | 10.749 | 2.437 | 185.108 | 133.168 | 6.626 | 11.540 | 1.591 | 19.155 | 126.925 | 6.203 | 11.266 | 2.547 | 120.160 |
| 8 | 113.355 | 4.967 | 10.647 | 2.888 | 298.107 | 134.383 | 6.804 | 11.592 | 1.236 | 6.780 | 132.905 | 6.641 | 11.528 | 1.209 | 14.174 | ||
| 10 | 114.407 | 4.984 | 10.696 | 2.164 | 153.069 | 131.382 | 6.631 | 11.462 | 1.483 | 26.411 | 128.820 | 6.443 | 11.350 | 1.539 | 32.792 | ||
| 12 | 117.801 | 5.253 | 10.854 | 2.692 | 326.651 | 128.201 | 6.430 | 11.323 | 1.408 | 26.498 | 136.753 | 7.039 | 11.694 | 1.168 | 4.854 | ||
| Methods | TaTS (ours) | TaTS (ours) | TaTS (ours) | N-BEATS | N-HiTS | TCN | ChatTime | GPT4MTS | ||||||||||
| + iTransformer | + PatchTST | + FiLM | ( 2020 ) | ( 2022 ) | ( 2018 ) | ( 2025 ) | ( 2024 ) | |||||||||||
| Datasets | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | MSE | MAE | ||
| Time-MMD ( 2024a ) | Agriculture | 6 | 0.067 | 0.184 | 0.066 | 0.171 | 0.087 | 0.205 | 2.436 | 1.210 | 2.274 | 1.134 | 3.957 | 1.770 | 0.502 | 0.441 | 0.197 | 0.292 |
| 8 | 0.094 | 0.210 | 0.096 | 0.217 | 0.110 | 0.223 | 3.671 | 1.446 | 1.575 | 0.866 | 5.078 | 1.919 | 0.510 | 0.449 | 0.352 | 0.426 | ||
| 10 | 0.122 | 0.251 | 0.126 | 0.260 | 0.146 | 0.249 | 3.078 | 1.416 | 1.441 | 0.910 | 4.464 | 1.866 | 0.505 | 0.445 | 0.358 | 0.413 | ||
| 12 | 0.153 | 0.271 | 0.166 | 0.292 | 0.196 | 0.328 | 3.883 | 1.759 | 2.119 | 1.217 | 3.173 | 1.633 | 0.517 | 0.452 | 0.404 | 0.444 | ||
| Models | PatchTST | DLinear | FiLM | |||||||||||||
| ( 2023 ) | ( 2023 ) | ( 2022a ) | ||||||||||||||
| Metric | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | MSE | MAE | RMSE | MAPE | MSPE | |
| Climate | Uni-modal | 1.111 | 0.846 | 1.052 | 3.898 | 442.010 | 0.969 | 0.801 | 0.983 | 1.881 | 59.681 | 1.123 | 0.829 | 1.059 | 2.647 | 128.421 |
| MM-TSFLib | 1.010 | 0.821 | 1.002 | 2.156 | 81.983 | 0.963 | 0.802 | 0.980 | 1.394 | 18.580 | 1.130 | 0.833 | 1.061 | 1.949 | 51.811 | |
| TaTS (ours) | 0.878 | 0.720 | 0.937 | 1.573 | 35.877 | 0.912 | 0.757 | 0.951 | 1.701 | 25.242 | 0.820 | 0.718 | 0.902 | 1.478 | 30.622 | |
| Promotion | 13.1% | 12.3% | 6.5% | 27.0% | 56.2% | 5.3% | 5.5% | 3.0% | -22.0% | -35.9% | 27.0% | 13.4% | 14.8% | 24.2% | 40.9% | |
| Text Embedding Method | Agriculture | Climate | Economy | Security | Social Good | Traffic |
| GPT-2 average pooling (prediction length 6) | 0.067 | 1.020 | 0.008 | 107.113 | 0.942 | 0.174 |
| LLaMA-3.2-1B sentence embeddings (prediction length 6) | 0.067 | 1.015 | 0.008 | 106.625 | 0.935 | 0.179 |
| GPT-2 average pooling (prediction length 12) | 0.153 | 1.033 | 0.008 | 114.754 | 1.045 | 0.213 |
| LLaMA-3.2-1B sentence embeddings (prediction length 12) | 0.152 | 1.021 | 0.008 | 114.219 | 1.025 | 0.209 |