Cold-start cybersecurity ontology population using information extraction with LSTM

• Published in: 2019 International Conference on Cyber Security for Emerging Technologies (CSET) pp. 1 - 6
• Main Authors: Gasmi, Houssem, Laval, Jannik, Bouras, Abdelaziz
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-10-2019
• Subjects: cybersecurity; named entity recognition; ontology learning; relation extraction
• DOI: 10.1109/CSET.2019.8904905

In this paper, we discuss how Long Short Time Memory (LSTM) neural networks can be applied to cyber security knowledge base population. Assuming we have an empty ontology that models the field of vulnerabilities description management using ontology concepts such as classes and properties, we want to populate it from online unstructured textual resources. More precisely, the task involves predicting instances of the classes in the ontology and the semantic relationship between them from a text describing a vulnerability in a software. As opposed to the statistical inference approach, we adopt a neural networks approach to predict the structure of the text. Given an input as a sequence of words, the model predicts the most likely classification of the words and extracts the relationship between the words that are relevant to the domain. The proposed system is decomposed into named entry recognition, relation extraction, ontology population. In this paper, we show how these tasks fit together and how they are implemented as unified framework.