Automated Influence Network Generation and the Node Parameter Sensitivity Analysis; Conference paper

摘要

An influence network is a directed graph extensively used for Effects-Based Operation. It contains nodes that represent events and links that encode causal relationships among events. It propagates the likelihood of each event through promotion or inhibition by its parents. As a subject matter expert often builds this network by hand, we helped simplify the influence network generation in Organization Risk Analyzer. The resulting influence network is generated from a multi-mode, multi-plex organizational network structure, and the generation scheme is based on assessing event flows and evaluating the factors on task management of the organization. To support the soundness of such network generation we provide sensitivity analysis of baseline probabilities, a major parameter of the model, by bootstrap sampling of the leaf nodes and propagating different levels of assigned parameters. Finally, we provide an example of analysis by utilizing the introduced generation method and a dataset from 1998 US embassy bombing in Kenya.