Automatic Detection of Metaphors in CyberSecurity Texts - Corpus and Model

Metaphorical expressions connect two conceptual domains, where one domain is comprehended and experienced in terms of the other. Using metaphors could greatly improve cybersecurity adoption, education, communication, and thinking. Metaphors can convey complex security concepts and aid in user comprehension, but their identification and interpretation can be challenging. We propose a method for automatically detecting metaphors in cybersecurity texts to address this challenge. First, we develop a cybersecurity metaphor corpus - CyMET using over 1000 cybersecurity texts scraped from the web. Using an adapted version of MIP for metaphor identification, we annotate the corpus and investigate the source domain identification within several of these cybersecurity texts. Next, we investigate the fine-tuning of the pre-trained NLP model, RoBERTa, for metaphor identification in cybersecurity texts. We explore optimizing and tuning the final model(cyBERTa). Finally, we briefly explore the potential of sonification and visualization techniques to aid in identifying and interpreting metaphors in cybersecurity texts. We create visual and aural representations for the metaphors and domains identified in our corpus. Our results demonstrate the effectiveness of our proposed method for automatic metaphor detection in cybersecurity texts. Our cybersecurity metaphor corpus - CyMET and cyBERTa provide valuable resources for further research. The potential of sonification and visualization techniques to aid in metaphor interpretation also opens up new avenues for cybersecurity communication and user education.

Embargo status: Restricted until 06/2024. To request the author grant access, click on the PDF link to the left.