.png){kind=icon}
.png){kind=icon}
Return
ContextMiner: Mining Contextual Features for Conceptualizing Knowledge in Security Texts
.png){kind=icon}
PRE
AI
DOI：10.1109/ACCESS.2022.3198944.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper presents ContextMiner, a novel natural language processing (NLP) framework to automatically capture contextual features for the purpose of extracting meaningful context-aware phrases from cybersecurity unstructured textual data. The framework utilizes basic attributes such as part-of-speech tagging, dependency parsing, and a domain-specific grammar to extract the contextual features. The effectiveness and applications of ContextMiner are evaluated and presented from two different perspectives: qualitative and quantitative. As for the qualitative analysis, our case studies show that the proposed framework is capable of retrieving additional contents from the given texts, both in a labeled and unlabeled setting, and thus building context-aware phrases in comparison with existing approaches. From a quantitative point of view, we evaluate ContextMiner as a pre-processing step to perform named entity recognition (NER). Our results show that ContextMiner reduces the corpus up to 70% while maintaining 85% of its relevant entities, with a small drop in the classification metrics. Finally, we explored the utilization of ContextMiner in the construction and reasoning of knowledge graphs.
Keywords:
Feature extraction
Computer security
Data mining
Syntactics
Natural language processing
Machine learning
Tagging
Dependency parsing
feature extraction
machine learning
natural language processing
word embeddings
