# Towards Early Warning of Real-World Vulnerabilities: An Evidence-Aware Decoupled Method for CVSS Vector Prediction Based on Unstructured Disclosures

DevFeed: [Towards Early Warning of Real-World Vulnerabilities: An Evidence-Aware Decoupled Method for CVSS Vector Prediction Based on Unstructured Disclosures](<https://devfeed.tech/articles/towards-early-warning-of-real-world-vulnerabilities-an-evidence-aware-decoupled-method-for-cvss-vector-prediction-based-on-unstructured-disclosures-118713.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06477>)

Author: Jianhong Ma, Wenhao Li, Xixing Li, Chenyu Yan, Haowen Chen

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [CVSS v4.0](<https://devfeed.tech/topics/cvss-v4-0.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [NVD](<https://devfeed.tech/topics/nvd.md>), [OSINT](<https://devfeed.tech/topics/osint.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [absolute](<https://devfeed.tech/tags/absolute.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [assessment](<https://devfeed.tech/tags/assessment.md>), [based](<https://devfeed.tech/tags/based.md>), [bench](<https://devfeed.tech/tags/bench.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [bert](<https://devfeed.tech/tags/bert.md>), [can](<https://devfeed.tech/tags/can.md>), [class](<https://devfeed.tech/tags/class.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cvss](<https://devfeed.tech/tags/cvss.md>), [disclosures](<https://devfeed.tech/tags/disclosures.md>), [early-warning](<https://devfeed.tech/tags/early-warning.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [vector](<https://devfeed.tech/tags/vector.md>), [vulnerabilities](<https://devfeed.tech/tags/vulnerabilities.md>), [warning](<https://devfeed.tech/tags/warning.md>), [world](<https://devfeed.tech/tags/world.md>)

## AI overview

VPre predicts CVSS vectors from noisy, incomplete vulnerability disclosures, with the benchmark showing earlier estimates than official NVD scoring. Its two-level method first screens reports and sentences for security evidence, then routes each of the eight CVSS metrics according to evidence availability: large-language-model reasoning for scattered evidence and lightweight supervised classification for explicit evidence. On an 851-report held-out split of VPre-Bench (5,581 reports total), it achieves mean absolute error 0.7239, reported as 44.7% below a reproduced BERT-Small-NVD early-transfer baseline. Across the benchmark, predictions precede official scoring by 38 days on average (observed range 21-60 days). In the stated 38:1 PARSE-Triage noise setting, STDU report screening reaches 0.98 precision, 1.00 recall and 0.99 F1. The abstract reports controlled ablations supporting both decoupling levels.

## Source excerpt

arXiv:2611.06477v1 Announce Type: new Abstract: Early assessment of Common Vulnerability Scoring System (CVSS) vectors from Open Source Intelligence (OSINT) can reduce the delay between initial public disclosure and official National Vulnerability Database (NVD) scoring. Existing predictors, trained mainly on standardized late-stage descriptions, degrade on noisy and incomplete pre-NVD reports. We introduce PARSE, a vulnerability-disclosure lifecycle model, and VPre, an early assessment framework built on two-level Evidence-Aware Decoupling. Decoupling Level I, Security-Evidence Screening, separates security-relevant evidence from noisy early reports at the report and sentence levels. Decoupling Level II, Evidence-Matched Routing, decouples the prediction task of the eight CVSS metrics into a scattered-evidence class and an explicit-evidence class based on validation-set measurements, matching each with a targeted predictor: large-language-model reasoning for the former, lightweight supervised classification for the latter. On the 851-report held-out test split of the 5,581-report VPre-Bench, VPre achieves a mean absolute error of 0.7239 -- 44.7% below the reproduced BERT-Small-NVD early-transfer baseline; paired timestamps across the full benchmark show an average 38-day lead before official scoring (observed range 21-60 days). Under the 38:1 PARSE-Triage noise setting, STDU's report-level screening reaches 0.98 precision, 1.00 recall, and 0.99 F1. Controlled ablations support the contributions of both decoupling levels.