Data-Driven Multi-Agent Email Generators

We present a method for generating realistic-looking emails for multi-agent simulations, motivated by the needs of cyber ranges. Our approach uses a distributed model of email threads which represents communication graphs learned from an email corpus. This explainable model uses a template-based generation system, also learned from an email corpus. Social network analysis measures are used to compare system-generated with human-generated email data. This generation system enables more rapid implementation of novel scenarios unrelated to the original corpus.