Christopher Bronk (Baker Institute, Rice University) and Derek Ruths (School of Computer Science, McGill University) will contribute two chapters to the book. The first one is Classification Inference & Event Attribution, and the second one is Policy Informatics – Reflections from User Observations.
Classification Inference & Event Attribution
A major objective of both academic and field research on violent groups is to devise a classification system that captures ideological, methodological, and behavioral relationships among them. In the last decade, development of new models incorporating thinking from international relations, computation, and area studies were actively solicited by security agencies in the U.S. and other countries. Our own interest in this topic led to some work on development of cross-disciplinary policy informatics work in the comparison and classification requisite for making higher-level observations about nature of these groups and even tactics that may be used to manage or mitigate their activities.
Lack of detailed information about the inner workings of violent groups, the large number of groups that exist, and the wide array of different types of groups are major obstacles to the construction of meaningful classifications. In this chapter, we present a methodology based on hierarchical clustering that uses entirely open information sources to construct a complete ontology over a set of violent groups. We formalize the History-Biased Event Participant Inference Problem identifying the violent groups most likely involved in a given event and then present a methodology that solves this problem. Our approach trained a naïve Bayesian model of violent group behavior using a data set of open-source events with known group involvement. This model is then used to identify groups whose behavioral model is most consistent with an event for which all participants (or culpable parties) are unknown.
In order to test our method, we used a data set provided by the Institute for the Study of Violent Groups (ISVG) which consisted of attributes of and responsible groups for over 28,000 violent events reported in open-source news sources. Collected events spanned a five year period from January 1, 2002 to December 31, 2007. We used our method to perform 10-fold cross-validations on this data set and determined that our method predicts true participants with over 80% accuracy. These results both validate the utility and performance of our method and reaffirm the ability for open-source intelligence to offer new insights into violent and clandestine group behavior.
We studied our method’s performance against data collected during the November 2008 Mumbai attacks. Using models constructed from the ISVG dataset and attributes of the attacks taken from news reports released directly following the incident, This case study, in particular, suggests that our method can provide valuable insights to those studying violent group behavior. We finally summarize the issues involved in the problem of policy informatics on rapidly evolving policy problems, in this case transnational terrorism, and the difficulties of joining generalized international relations method with computation and complex area studies knowledge.
A major objective of both academic and field research on violent groups is to devise a classification system that captures ideological, methodological, and behavioral relationships among them. In the last decade, development of new models incorporating thinking from international relations, computation, and area studies were actively solicited by security agencies in the U.S. and other countries. Our own interest in this topic led to some work on development of cross-disciplinary policy informatics work in the comparison and classification requisite for making higher-level observations about nature of these groups and even tactics that may be used to manage or mitigate their activities.
Lack of detailed information about the inner workings of violent groups, the large number of groups that exist, and the wide array of different types of groups are major obstacles to the construction of meaningful classifications. In this chapter, we present a methodology based on hierarchical clustering that uses entirely open information sources to construct a complete ontology over a set of violent groups. We formalize the History-Biased Event Participant Inference Problem identifying the violent groups most likely involved in a given event and then present a methodology that solves this problem. Our approach trained a naïve Bayesian model of violent group behavior using a data set of open-source events with known group involvement. This model is then used to identify groups whose behavioral model is most consistent with an event for which all participants (or culpable parties) are unknown.
In order to test our method, we used a data set provided by the Institute for the Study of Violent Groups (ISVG) which consisted of attributes of and responsible groups for over 28,000 violent events reported in open-source news sources. Collected events spanned a five year period from January 1, 2002 to December 31, 2007. We used our method to perform 10-fold cross-validations on this data set and determined that our method predicts true participants with over 80% accuracy. These results both validate the utility and performance of our method and reaffirm the ability for open-source intelligence to offer new insights into violent and clandestine group behavior.
We studied our method’s performance against data collected during the November 2008 Mumbai attacks. Using models constructed from the ISVG dataset and attributes of the attacks taken from news reports released directly following the incident, This case study, in particular, suggests that our method can provide valuable insights to those studying violent group behavior. We finally summarize the issues involved in the problem of policy informatics on rapidly evolving policy problems, in this case transnational terrorism, and the difficulties of joining generalized international relations method with computation and complex area studies knowledge.
The authors wish to enter a chapter on the practical lessons of policy informatics research, drawing upon their experience in the last decade. This chapter would provide views on several areas, including: (1) the linkage between scholarship and practitioner input on the research agenda of computing and politics; (2) the problem of discontinuity in data, produced by dramatic shifts in global social and political organization, enabled by globalization and the Internet; and (3) the issues of overlapping and orthogonal relationships regarding expertise and input in which computational methodology and political subject matter merge.
Policy Informatics – Reflections from User Observations
Presented are our lessons learned from several projects. Our primary focus is in the domain of international relations (IR), however, we accept that barriers and sovereignty are concepts grossly redefined by the explosion in digital connectivity around the globe in the past two decades. We differentiate here between policy informatics, which we would generally argue to be the application of computational or algorithmic methods to policy or political research and studies in information policy, which may take form in more conventional quantitative or qualitative form. We assert that, research programs that apply computational method to better understand political phenomena are a rapidly developing area of scholarship. That said, significant obstacles exist in the study of political behavior by algorithmic rather than statistical process. American political science has dedicated significant effort to the understanding of IR phenomena through the study of data regarding the actions of states via statistical means. Two major developments challenge this approach to IR scholarship, the creation of the Internet and user-driven communications platforms (i.e. Twitter, Facebook and YouTube), and the capacity to rapidly develop software designed to study new sets of structured and unstructured data.
Our own experience draws from several subjects: the institutional Wiki platform at the U.S. Department of State, Diplopedia; transnational terror group actions and behavior; Internet filtering activities; and the aggregation of attention in political discourse and crisis on the Twitter social media platform. These our major collaborative projects have taught us important and painful lessons on the study of revolution, in this case an Information Revolution, which we believe to be ongoing. We aim to illustrate a path to policy informatics scholarship that we have pursued and where we believe interesting work is to be found.
Anand Desai, of the John Glenn School of Public Affairs, at the Ohio State University will write a chapter on The Evidentiary Basis of Policy Inquiry.
