TimelyRAG: Semantic-Temporal Hybrid Retrieval for Time-Critical Question Answering with Evolving Documents
This is the official implementation of TimelyRAG
| Name | Domain | #Docs | #Queries | Temporal Type |
|---|---|---|---|---|
| Law | Legal Regulations | 30,000 | 3,000 | Overlapping-Evolving |
| University | Academic Policies | 30,000 | 3,000 | Overlapping-Evolving |
| Company Policies | Corporate Guidelines | 30,000 | 3,000 | Overlapping-Evolving |
| Terms of Service | Service Policies | 30,000 | 3,000 | Overlapping-Evolving |
- Node.js: 16.13.2+
- Anaconda 4 or Miniconda 3
- Python 3.11.5 (Recommend Anaconda)
- Ubuntu 18.04.6 LTS
- pytorch >= 2.1.2
All hyperparameters and paths are managed in config.py.
- Parameter options
--exp-dir: Experiment output directory (string)
--gpu: GPU device id (int)
--datasets: Comma-separated JSONL dataset paths (list)