Efficient Construction of Dynamic Meta-Networks

Our objective is to enable military analysts to rapidly produce intelligence results regarding adversarial groups and so diverse types of covert networks, and to improve the quality of the underlying analyses. We focus, on the data entry and cleaning problems. We propose a mixed-initiative (human+machine) approach involving text mining, visual data entry and cleaning aids to improve multi-source data collection, fusion and cleaning that enable the rapid construction of network data yet stay attuned to the level and source of uncertainty and the mission focus. The proposed approach supports meta-data extraction, incremental data collection and value added assessment. We utilize text mining, machine learning, standard thesauri and import wizards to rapidly extract from raw texts details on networks connecting who, what, where, how and why, attributes of the nodes and links, statistical techniques for inferring missing data, and network assessment capabilities to assess the impact of coding choices on resultant network assessments. To reduce time from data collection to decision we employ mission driven data model refinement and extraction, uncertainty tracking, general, DIME/PMESII and adversarial ontologies, topic modeling for thesauri construction, multi-source data entry, network fusion, and linkages to novel collection technology such as crowd sourcing and cell-phones.