ST-Bench: Spatial-Temporal Reasoning Benchmark

ST-Bench is a comprehensive benchmark dataset for training and evaluating spatial-temporal reasoning capabilities in large language models. It includes data with raw time series, text descriptions, and image visualizations.

📊 Dataset Overview

Default Data (with time_series key)

Text-based Data (text description only)

Image-based Data (with image reference)

Note: The visualization images referenced in Image-based data are not included in this release due to size constraints.

📁 Directory Structure

ST-Bench/
├── ST-Align/                    # Alignment data
│   ├── alignment.jsonl
│   ├── alignment_train.jsonl
│   └── alignment_test.jsonl
│
├── ST-Causal/                   # Causal reasoning
│   └── causal.jsonl
│
├── ST-SFT/                      # Supervised Fine-Tuning (Default)
├── ST-CoT/                      # Chain-of-Thought (Default)
├── ST-RL/                       # Reinforcement Learning (Default)
├── ST-Test/                     # Test Set (Default)
│
├── ST-SFT-Text/                 # Supervised Fine-Tuning (Text)
├── ST-CoT-Text/                 # Chain-of-Thought (Text)
├── ST-RL-Text/                  # Reinforcement Learning (Text)
├── ST-Test-Text/                # Test Set (Text)
│
├── ST-CoT-Image/                # Chain-of-Thought (Image)
└── ST-RL-Image/                 # Reinforcement Learning (Image)

Each reasoning folder contains:

• correlation_*.jsonl - Correlation reasoning tasks
• entity_*.jsonl - Entity reasoning tasks
• etiological_*.jsonl - Etiological (cause-effect) reasoning tasks
• forecasting_*.jsonl - Forecasting tasks

🎯 Task Types

The reasoning tasks cover four main categories:

1. Correlation: Understanding relationships between time series variables
2. Entity: Identifying and reasoning about entities in temporal data
3. Etiological: Cause-effect reasoning in temporal sequences
4. Forecasting: Predicting future values based on historical patterns

📖 Usage

from datasets import load_dataset

# Load default data (with time_series key)
cot_data = load_dataset("Time-HD-Anonymous/ST-Bench", "ST-CoT")

# Load text-only data
cot_text = load_dataset("Time-HD-Anonymous/ST-Bench", "ST-CoT-Text")

# Load image-based data
cot_image = load_dataset("Time-HD-Anonymous/ST-Bench", "ST-CoT-Image")

# Load alignment data
align_data = load_dataset("Time-HD-Anonymous/ST-Bench", "ST-Align")

📜 License

This dataset is released under the Apache 2.0 License.

📝 Citation

If you use this dataset in your research, please cite:

@misc{stbench2025,
  title={ST-Bench: A Benchmark for Spatial-Temporal Reasoning},
  author={Anonymous},
  year={2025},
  publisher={Hugging Face}
}