---
dataset_name: embedding-cve-nvd-dataset
language:
  - en
license: mit
tags:
  - cybersecurity
  - cve
  - embeddings
  - nvd
pretty_name: CVE NVD Embedding Dataset
task_categories:
  - text-retrieval
task_ids:
  - document-retrieval
size_categories:
  - 100K<datasets<1M
---

# CVE NVD Embedding Dataset

This dataset contains the processed CVE/NVD corpus that was used with the `rag_mixedbread` pipeline.
It bundles:

- `cve_corpus.jsonl` (~700 MB): each line is a JSON object with `cve_id`, `title`, `description`, `cvss`, `vendors`, and the pre-computed text chunk that feeds the embedding model.
- `decomposed_query_results.json` (63 KB): a dictionary of exemplar queries, decomposed sub-questions, and the retrieved doc IDs used for quality checks.

## Generation pipeline
1. Raw CVE/NVD feeds were normalized via `prepare_cve_corpus.py`.
2. Fields were concatenated and deduplicated into retrieval-ready passages.
3. The resulting corpus was used to build MixedBread vector indexes for the RAG workflow.

## Usage
You can stream the JSONL file and index it with any vector database:

```python
import json
from datasets import load_dataset

ds = load_dataset("Kushalkhemka/embedding-cve-nvd-dataset", split="train")
for row in ds:
    payload = json.loads(row["text"])  # each row is a JSON line
    # index payload["chunk"] into your vector store
```

The `decomposed_query_results.json` file is useful for evaluation—each entry has the original user question, the decomposed sub-queries, and the reference CVE IDs that should match during retrieval.

## License
MIT. Please respect the original NVD data terms when redistributing.