Towards explainability framework for cybersecurity domain: a case study using NER

This work presents some activities performed within the SoBigData.it research project, with the purpose of ad- dressing some critical challenges of Named Entity Recognition (NER) in the cybersecurity domain, by introducing a specific explainability framework. In detail, explainability techniques based on the Captum library have been experimented to deepen the analysis of model behaviour, revealing critical insights into feature importance and layer-wise contributions for entities like Malware, Vulnerability, and Indicators, showing a promising path for the adoption of similar approaches.