To address the subjectivity and inefficiency of manual common vulnerability scoring system (CVSS) assessment and the limitations of existing automated methods in capturing inter-metric relationships, a CVSS metrics prediction model of vulnerability based on a multi-gate shared and private mixture of experts (MSPMoE) is proposed. In the model, a shared-private expert mechanism is used to employed enyto jointly leverage shared and unique features across different metrics. Firstly, the DistilBERT model is fine-tuned on a collected vulnerability description corpus as the feature encoder. Then, a dynamic feature extraction module is designed to adaptively select optimal feature representation strategies. Finally, the MSPMoE classifier is adopted, where shared experts are used to capture common patterns across metrics, while private experts are used to extract metric-specific features. A gating network is used to dynamically weight each expert’s contribution, to balance shared and private knowledge for accurate CVSS metrics prediction. Experiments show that the proposed model outperforms state-of-the-art methods on seven CVSS metrics, achieving an average accuracy of 83.32% across all eight metrics.
此外,多项研究持续探索多元化的预训练模型以优化预测效果,Shi等[16]引入XLNet以利用其排列语言建模机制捕捉更丰富的语义特征;A等[17]与Isogai等[18]则探索了GPT系列模型在CVSS预测中的生成式潜力;Pourbehzadi等[19]探索融合解释主成分分析和支持向量回归的漏洞自动化评分框架;Ghosh等[20]提出CVE-LLM模型,将大语言模型应用于漏洞自动化评估;Khan等[21]结合SMOTE与欠采样技术解决类别不平衡问题,为CVSS预测提供了一个轻量化的框架。当前CVSS度量预测模型尚未有效兼顾任务关联性与独立性。单任务模型结构清晰但忽略度量关联,难以利用度量共享特征;多任务模型虽能捕捉度量共享特征,却不能区分不同度量的语义与预测维度差异,无法充分学习度量私有特征。为此,彭真[22]基于DistilBERT模型,引入多门混合专家(Multi-Gate Mixture of Experts, MMoE)架构[23]预测各CVSS度量。该方法通过多个共享专家网络与度量专属门控网络,在一定程度上利用了各度量的私有特征。然而其专家网络本身为共享结构,仍难以充分捕获各度量的私有特征。
因此,提出一种基于多门共享—私有混合专家(Multi-Gate Shared and Private Mixture of Experts, MSPMoE)的漏洞CVSS度量预测模型,通过共享与私有特征的双路径学习机制,有效捕捉度量间共性并保留其私有特征。模型包含3个核心组件:经领域语料微调的DistilBERT特征编码器,增强对安全术语的语义表征;自适应特征提取器,动态捕捉深层上下文特征;以及MSPMoE分类器,通过共享专家与私有专家的协同机制与门控网络,实现共享知识与私有知识的自适应融合,全面提升预测性能。
ROSTAMIA, MOHAMMADIM, KARIMIPOURH. Reliability assessment of cyber-physical power systems considering the impact of predicted cyber vulnerabilities[J]. International Journal of Electrical Power & Energy Systems, 2023,147:No.108892.
[6]
AKSUM U, DILEKM H, TATLıE İ, et al. A quantitative CVSS-based cyber security risk assessment methodology for IT systems[C]//Proceedings of the 2017 International Carnahan Conference on Security Technology. Piscataway,USA: IEEE, 2017. DOI:10.1109/CCST.2017.8167819 .
[7]
YAMAMOTOY, MIYAMOTOD, NAKAYAMAM. Text-mining approach for estimating vulnerability score[C]//Proceedings of the 2015 4th International Workshop on Building Analysis Datasets and Gathering Experience Returns for Security. Piscataway, USA: IEEE, 2015:67-73.
[8]
KHAZAEIA, GHASEMZADEHM, DERHAMIV. An automatic method for CVSS score prediction using vulnerabilities description[J]. Journal of Intelligent & Fuzzy Systems, 2015,30(1):89-96.
[9]
SPANOSG, ANGELISL, TOLOUDISD. Assessment of vulnerability severity using text mining[C]//Proceedings of the 21st Pan-Hellenic Conference on Informatics. New York, USA: ACM, 2017:No.49.
[10]
HANZ B, LIX H, XINGZ C, et al. Learning to predict severity of software vulnerability using only vulnerability description[C]//Proceedings of the 2017 IEEE International Conference on Software Maintenance and Evolution. Piscataway, USA: IEEE, 2017:125-136.
[11]
SAHINS E, TOSUNA. A conceptual replication on predicting the severity of software vulnerabilities[C]//Proceedings of the Evaluation and Assessment on Software Engineering. New York, USA: ACM, 2019:244-250.
[12]
GONGX, XINGZ C, LIX H, et al. Joint prediction of multiple vulnerability characteristics through multi-task learning[C]//Proceedings of the 2019 24th International Conference on Engineering of Complex Computer Systems. Piscataway, USA: IEEE, 2019:31-40.
[13]
SHAHIDM R, DEBARH. CVSS-BERT: explainable natural language processing to determine the severity of a computer security vulnerability from its description[C]//Proceedings of the 2021 20th IEEE International Conference on Machine Learning and Applications. Piscataway, USA: IEEE, 2021: 1600-1607.
[14]
KÜHNP, RELKED N, REUTERC. Common vulnerability scoring system prediction based on open source intelligence information sources[J]. Computers & Security, 2023,131:No.103286.
[15]
COSTAJ C, ROXOT, SEQUEIROSJ B F, et al. Predicting CVSS metric via description interpretation[J]. IEEE Access, 2022,10:59125-59134.
[16]
SHIF, KAIS F, ZHENGJ H, et al. XLNet-based prediction model for CVSS metric values[J]. Applied Sciences, 2022,12(18):No. 8983.
[17]
A M, KOTAK, BABUA S, et al. CVE severity prediction from vulnerability description-a deep learning approach[J]. Procedia Computer Science, 2024,235:3105-3117.
[18]
ISOGAIS, OGATAS, KASHIWAY, et al. Toward extracting learning pattern: a comparative study of GPT-4o-mini and BERT models in predicting CVSS base vectors[C]//Proceedings of the 2024 IEEE 35th International Symposium on Software Reliability Engineering Workshops. Piscataway, USA: IEEE, 2024:127-134.
[19]
POURBEHZADIM, JAVIDIG, HOWELLC J, et al. Enhanced (cyber) situational awareness: using interpretable principal component analysis (iPCA) to automate vulnerability severity scoring[J]. Decision Support Systems, 2024,186:No.114308.
[20]
GHOSHR, FARRIO, VON STOCKHAUSENH M, et al. CVE-LLM: automatic vulnerability evaluation in medical device industry using large language models[J]. ArXiv e-Prints, 2024:arXiv:
[21]
KHANT, MALIKM, AZIZZ, et al. Threat and vulnerability classification using NLP and machine learning techniques on text-based security data[J]. Journal of Computers and Intelligent Systems,2025,3(2):139-152.
[22]
彭真.配电网信息物理系统网络安全风险评估技术研究[D].北京:北京交通大学,2024:1-81.
[23]
SHAZEERN, MIRHOSEINIA, MAZIARZK, et al. Outrageously large neural networks: the sparsely-gated mixture-of-experts layer [DB/OL]. (2017-01-23) [2025-11-10].
[24]
FaceHugging.The AI community building the future [DB/OL]. (2023-12-23)[2025-11-10].
[25]
MIKOLOVT, CHENK, CORRADOG, et al. Efficient estimation of word representations in vector space [DB/OL]. (2013-01-16)[2025-11-10].
[26]
DEVLINJ, CHANGM W, LEEK, et al. BERT: pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg, USA: ACL,2019:4171-4186.
[27]
CHIP H, CHUNGP H, WUT H, et al. Audio albert: a lite bert for self-supervised learning of audio representation[C]//Proceedings of the 2021 IEEE Spoken Language Technology Workshop. Piscataway, USA: IEEE, 2021:344-350.