Predicting organizational cybersecurity risk: a deep learning approach

This paper builds a tool called HackER that reads posts on hacker forums to spot software exploits and figure out which kinds of organizations they are aimed at. Using a deep learning model (a BiLSTM), it predicts the type of business an exploit targets, outperforming standard machine learning baselines with an F1-score of about 80%. The goal is to give security analysts an early warning about who attackers are likely to hit.