Add dataset card for D-SCAN
#2
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- text-classification
|
| 4 |
+
tags:
|
| 5 |
+
- rag
|
| 6 |
+
- security
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# D-SCAN Dataset
|
| 10 |
+
|
| 11 |
+
This dataset contains questions and retrieved documents for detecting RAG poisoning attacks, as described in [When Context Bites: Detecting RAG Poisoning via Document-Level Attention Collapse](https://huggingface.co/papers/2608.06947).
|
| 12 |
+
|
| 13 |
+
The data is used to train classifiers to distinguish between clean and poisoned retrieved documents in retrieval-augmented generation (RAG) systems. See the [GitHub repository](https://github.com/yingtaoren/D-Scan.git) for details on data collection, feature computation, and classifier training.
|
| 14 |
+
|
| 15 |
+
Code: https://github.com/yingtaoren/D-Scan.git
|