Tuning Automatic Summarization for Incident Report Visualization

Abstract

We present a machine learning approach to improve the accuracy of summarized incident report visualizations for cyber security. We extend a recent incident report summarization method by training a Bayesian hierarchical model to optimize the summarization algorithm's weights. We also train a flat model and a neural network as alternative models to compare against our hierarchical model. Summaries generated by our hierarchical model achieve higher accuracy than the other methods, with an AUC 0.2 higher than the unweighted method while achieving comparable summarization size. We further demonstrate that visualizations of the hierarchical model's summaries are at least as useful the unweighted method's summaries, and possibly more useful.