D3.3 Implement Hierarchy Reconstructing Methods

Terrorism has its own frame and structure. As all organizations that consist of many individuals and conduct several tasks, the actions of terrorists are driven by a hierarchical background. However, in several cases, this hierarchy is hidden and builds up in a self-organized way. For traditional observation techniques, this organization seems to be wide spread, unstructured and loosely connected. Here comes Social Network Analysis into an important role: collecting small pieces of information from huge amount of data results in a holistic picture, where – if data allows it – the unseen hierarchical skeleton can be revealed.

The WP3 “Social Network Analysis”, aims to provide tools and software solutions for handling relational data. This report presents the component for “Implement hierarchy reconstructing methods”, which results from Task 3.3 and describes implemented algorithms for revealing hierarchical structures from flat dataset, and is a part of the SNA module of RED-Alert project. The component constructs new networks from input data: either from co-occurrence statistics or from directed networks containing loops. Furthermore, quantitative measures are calculated for characterizing the similarity of any network to an ideal hierarchical structure. This report presents a short summary about the field experiences of Law Enforcement Agencies on the formation and properties of terrorist organizations. Further research and enhancement of the presented tools will be based on these guidelines. The core modules for hierarchy reconstruction are described with the appropriate theoretical background.

According to the RED-Alert System Architecture (D1.7, Fig. 20), the result of this deliverable as part of the WP3 will function as a batch processing, middle layer between the processors of raw input (NLP and SMA) and the output (CEP and Visualization). Because SNA works with topological and statistical information, it does not need to have access to raw data, all input can be fully anonymized. Only the topological (relational) information must be kept from the original source during anonymization. In several cases, this requires pseudonym data instead of full (aggregated and randomized) anonymization. In cases of pseudonym data, the key for linking back to real persons, will be kept separately, the SNA analysis will have no access to the keys, which renders the pseudonym data to be practically anonymous (reverse engineering would require disproportional efforts). The results of this task are expected to be utilized in the Visualization tool and in NLP. The latter can benefit from taxonomy/ontology creation, when the presented algorithms are applied to linguistic entities.

Read the full deliverable: D3 3 Hierarchical models FINAL