Improved Method for Cybersecurity Entity Extraction
Author Information
Author(s): Zhang Ke, Wang Yunpeng, Li Ou, Hao Sirui, He Junjiang, Lan Xiaolong, Yang Jinneng, Ye Yang
Primary Institution: Nuclear Power Institute of China, Chengdu, China
Hypothesis
Can an improved self-training-based distant label denoising method enhance the quality of cybersecurity entity extraction?
Conclusion
The proposed method effectively improves the performance of several state-of-the-art named entity recognition models in the cybersecurity domain.
Supporting Evidence
- The proposed method achieved a 3.5% improvement for the Vendor class and a 3.35% improvement for the Product class in F1 scores.
- Experimental results demonstrated that the quality of the distant labels generated was high.
- The constrained self-training algorithm effectively improved the performance of several state-of-the-art NER models.
Takeaway
This study shows a new way to help computers understand cybersecurity information better by using smart training methods.
Methodology
The study involved creating domain dictionaries, generating distant labels using a reverse maximum matching algorithm, and applying an improved self-training algorithm to enhance model performance.
Potential Biases
The reliance on distant supervision may introduce noise in the labels, affecting model performance.
Limitations
The study only utilized two cybersecurity lexicons, which may limit the coverage of domain-specific terms.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website