pvv5385/cve-embeddings

CVE Flat Table

A flattened, tabular export of the official CVE List v5 (CVEProject/cvelistV5), one row per CVE record, built for embedding / semantic-search / classification experiments.

Source data is public-domain CVE Program data (CVE Record Format 5.x). This export was generated on 2026-09-17 from the latest daily baseline release.

Fields

• cve_id — CVE identifier
• state — record state (e.g. PUBLISHED, REJECTED)
• description — English-language vulnerability description (from the CNA container)
• cwe_ids — semicolon-separated CWE identifiers
• cvss_v2 / cvss_v3 / cvss_v4 — base scores, where reported
• severity — base severity, where reported
• vendors / products — semicolon-separated affected vendor/product names
• date_published / date_updated — CVE Program metadata timestamps

Embeddings

cve_embeddings.npy holds one dense vector per row (float32, L2-normalized), generated with BAAI/bge-large-en-v1.5 over description | CWE: ... | Products: .... Row order matches cve_embeddings.ids.parquet (cve_id column) — join on that, not on row order against cve_flat.parquet alone, since future updates may append/reorder rows. Load with:

import numpy as np, pandas as pd
emb = np.load("cve_embeddings.npy")
ids = pd.read_parquet("cve_embeddings.ids.parquet")

Notes

• Only records with a non-empty description are included.
• vendors/products reflect whatever the CNA reported; some records use literal "n/a" values rather than leaving the field empty.
• This is a raw extraction, not an official CVE Program artifact. Refer to the CVE List v5 repo for the authoritative source and terms of use.