Loading…
Context-Aware Attended-over Distributed Specificity for Information Extraction in Cybersecurity
Cybersecurity can benefit substantially from a timely discovery of cyber threats in part taking advantage of information extraction techniques for capturing vulnerabilities that are communicated via textual channels. One of the latest techniques was attended-over distributed specificity (ADS) for te...
Saved in:
Main Authors: | , , , |
---|---|
Format: | Conference Proceeding |
Language: | English |
Subjects: | |
Online Access: | Request full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Cybersecurity can benefit substantially from a timely discovery of cyber threats in part taking advantage of information extraction techniques for capturing vulnerabilities that are communicated via textual channels. One of the latest techniques was attended-over distributed specificity (ADS) for term extraction, which was tested on a cybersecurity dataset for this use case called Cyber ATE. However, in the absence of provisioning context in the dataset, the detection of terms could still occasionally fail. Thus, in our present work, we supplement CyberATE by including context phrases surrounding term candidates with varying window sizes. Then, we designed and evaluated new architectures for context-aware ATE. Our new architectures considerably beat the previous state-of-the-art, by as much as 12.48% F1-score. Further, we also show that providing context helps in attaining better classification, especially in improving recall, reaching as high as 5 % improvement. This is highly important especially in discovering new vulnerabilities in cybersecurity, since false negatives could lead to undiscovered vulnerabilities. Efficient and robust discovery of critical information pertaining to new cyber threats can facilitate and speedup the process to address vulnerabilities, well before any cyberattacks aim the target system or operation. |
---|---|
ISSN: | 2644-3163 |
DOI: | 10.1109/IEMCON56893.2022.9946567 |