Constructing and Analyzing Uncertain Social Networks from Unstructured Textual Data推荐.pdfVIP

Constructing and Analyzing Uncertain Social Networks from Unstructured Textual Data Fredrik Johansson and Pontus Svenson Abstract Social network analysis and link diagrams are popular tools among in- telligence analysts for analyzing and understanding criminal and terrorist organi- zations. A bottleneck in the use of such techniques is the manual effort needed to create the network to analyze from available source information. We describe how text mining techniques can be used for extraction of named entities and the relations among them, in order to enable automatic construction of networks from unstruc- tured text. Since the text mining techniques used, viz. algorithms for named entity recognition and relation extraction, are not perfect, we also describe a method for incorporating information about uncertainty when constructing the networks and when doing the social network analysis. The presented approach is applied on text documents describing terrorist activities in Indonesia. Keywords NLP · Relation extraction · SNA · Social network analysis · Text mining 1 Introduction Visualization of network structures has long been an important tool for police and intelligence analysts who are investigating criminal or terrorist networks (also re- ferred to as dark networks [1, 2]). However, social network analysis (SNA) is a rich research field that contains more than just visualization. It has in recent years also become increasingly popular among analysts and investigators to use SNA-related techniques such as node centrality measures and community detection algorithms. An example of how such techniques can be applied is to use them when deciding whom to bring in for questioning in criminal investigations [3]. In [4] it is discussed how SNA metrics can be used to judge which members of terrorist networks that are F. Johansson ( ) · P. Svenson B Swedish Defence Research Agency (FOI), SE-164 90, Stockholm, Sweden e-