MultiScale Modeling of Radical and Counter-Radical Islamic Organizations

Nyunsu Kim, Sukru Tikves,Zheng Wang, Hasan DavulcuArizona State University, USAfnkim,sukru,zwang,hdavulcug@asu.eduZeki ErdemTurkish NationalScience Foundation, Turkeyzeki.erdem@tubitak.gov.trJonathan Githens-MazerDepartment of PoliticsUniversity of Exeter, Exeter, UKJ.Githens-Mazer@exeter.ac.ukABSTRACTIn this paper we utilize an ecient sparse inversecovariance matrix (precision matrix) estimation tech-nique to identify a set of highly correlated discrimina-tive perspectives between radical and counter-radicalIslamic organizations. We develop a ranking systemthat utilizes ranked perspectives to map Islamic or-ganizations on a set of socio-cultural, political andbehavioral scales based on their web corpus. We cre-ate a gold standard ranking of these organizationsthrough an expertise elicitation tool. We computeexpert-to-expert agreements, and present experimen-tal results comparing the performance of the QUICbased scaling system to another baseline method for26 Islamic organizations in UK. The QUIC basedalgorithm not only outperforms the baseline method,but it is also the only system that consistently per-forms at area expert-level accuracies for all scales.I INTRODUCTIONWe propose a multi-scaling based methodology thatrepresents an important step change in how we mightobserve and analyze radical and counter-radical Is-lamic groups in any speci c region. Rather thanplacing external forms of analysis that color and tau-tological de ne what is ‘radical’ or not, we proposea more ontologically oriented approach. We seek todevelop a methodology to allow the orientations ofthese groups to de ne themselves via their own dis-course within their own universe and understandingof actions, rather than an external and potentiallypoorly calibrated analysis of what constitutes radical.Without this kind of fundamental reorientation to re-search of religiously or politically inspired groups, weget the poor assumption based analysis that (incor-rectly) predicts and champions ill-de ned relation-ships between certain religious or political sects andviolence, for example. With our reorientation of ap-proach, we are more fundamentally able to exam-ine such relationships in a way that should allow re-searchers to take other kinds of nuance and under-standing into account.In the case of Islamic social movements, EdwardSaid [1] observed, the boundary between politicalrhetoric and scholarship concerning Islam is oftenblurred. The problem is particularly acute when itcomes to the study of violent forms of political Islamand others deemed to be potentially violent. Much ofthe analytic and policy oriented literature relies on bi-nary distinctions such as \radical/moderate", \mod-ern/traditional", \conservative/progressive" etc. Bi-nary models map enormous diversity into ill-de nedcategories that often measure a mix of attitudesabout democracy, secularism, attitudes about theWest and proclivity to violence.This paper explores these problems and presents amulti-scale model based on more precise and objec-tive criteria that can be used to evaluate and comparemovements in diverse cultural, historical, and polit-ical contexts and how they change over time. Theanalysis and modeling e orts build on previous stud-ies [2] of change oriented social movements and use acombination of ethnographic, discourse analysis, andcomputational methods as well as a case study in-volving 26 Islamic groups from the United Kingdom(UK). The model presented here aims to broaden thebase of discussion and analysis, recapture and buildupon previous observations, and establish a generalframework within which critically needed compara-tive studies looking at both violent and non-violentgroups can be conducted.One of the fundamental issues with interpretative andqualitative data collection and analysis of groups andsocial movements has been the researchers’ bias whileconducting the research. Goertz [3] makes the cru-cial point that, in their enthusiasm for reifying com-plex sociological, cultural or political concepts, the-orists and empiricists often focus too much on whata concept is, rather than on identifying the concepton a continuum, in order to assess when a concept ispresent versus when it is absent.In the social sciences, scaling is the process of mea-suring and ordering actors (subjects) with respect toquantitative attributes or traits (items). In this pa-per, we present graphical tools and computational