FREE ELECTRONIC LIBRARY - Abstract, dissertation, book

Pages:   || 2 |

«Abstract Stochastic Opponent Modeling Agents (SOMA) have been proposed as a paradigm for reasoning about cultural groups, terror groups, and other ...»

-- [ Page 1 ] --

The SOMA Terror Organization Portal (STOP):

social network and analytic tools for the real-time

analysis of terror groups

Amy Sliva†, V.S. Subrahmanian†, Vanina Martinez†, and Gerardo I. Simari†

{asliva,vs,mvm,gisimari,asliva}@cs.umd.edu, Lab for Computational Cultural

Dynamics, Institute for Advanced Computer Studies, University of Maryland, College

Park, MD 20742, USA


Stochastic Opponent Modeling Agents (SOMA) have been proposed as a paradigm for reasoning about cultural groups, terror groups, and other socioeconomic-political-military organizations worldwide. In this paper, we describe the SOMA Terror Organization Portal (STOP). STOP provides a single point of contact through which analysts may access data about terror groups world wide. In order to analyze this data, SOMA provides three major components: the SOMA Extraction Engine (SEE), the SOMA Adversarial Forecast Engine (SAFE), and the SOMA Analyst NEtwork (SANE) that allows analysts to find other analysts doing similar work, share findings with them, and let consensus findings emerge. This paper describes the STOP framework.

1 Introduction Stochastic Opponent Modeling Agents introduced in [1,2,3] were introduced as a paradigm for reasoning about any group G in the world, irrespective of whether the group is a terror group, a social organization, a political party, a religious group, a militia, or an economic organization. SOMA-rules have been used to encode the behavior of players in the Afghan drug economy [4] as well as various tribes along the Pakistan-Afghanistan border [5], as well as terror groups such as Hezbollah [6]. Continuing studies are expected to track over 300 terror groups by end 2008.

The study of terror groups in the national security community is hampered by many major problems.

1. Lack of timely, accurate data. Data about the terror groups is collected manually. This causes any data to be either incomplete or out of date (usually both). One of the best known collections of data about terror groups is the “Minorities at Risk Organizational Behavior” (MAROB) data set [7,8] which tracks approximately 400 properties of ethnopolitical groups representing 284 ethnic groups worldwide. As mentioned in [5], such manual data collections are incomplete (because the manual coders of this data are only able to process certain numbers of articles in a limited number of languages) and coarse grained (because the variables are classified into categories rather than specific numbers – for instance, the number of deaths caused by a group might be classified as “none”, “low”, “medium” or “high” instead of giving a reasonably accurate estimate). Moreover, they are usually out of date (Minorities at Risk coding is currently complete only until 2003), leaving the data to be “non-actionable”.

2. Lack of behavioral models. Analysts today are forced to make up behavioral models in a slow and painstaking manner, The SOMA Extraction engine (SEE), a portion of STOP, extracts behavioral rules automatically from the MAROB data. To date, it has extracted rules on approximately 23 groups with another 2 or 3 groups being analyzed and completed each week.

3. Lack of a social network. Analysts are often unaware of what other analysts have found useful. There is no computational mechanism we have seen to date that allows analysts to share their experiences and forecasts for a particular group. In fact, analysts are often in the dark about who else is looking at a certain group – either from the same, or different viewpoint as themselves. The SOMA Analyst NEtwork (SANE) provides a social networking framework within which analysts can browse data about the groups of interest, rate the accuracy of the data, browse rules extracted by SEE, rate the rules extracted by SEE, and see the predictions produced by SAFE.

4. Lack of a forecast engine. The SOMA Adversarial Forecast Engine (SAFE) uses a rich probabilistic foundation with no unwarranted independence assumptions in order to forecast the most probable sets of actions that a terror group might take in a given situation. Using SAFE, analysts can analyze the current situation and/or hypothesize new situations based on possible actions the US might be contemplating.

We will not be discussing problem (1) above in much detail in this paper because our CARA architecture described in [9] already addresses many of the problems of dealing with real time data. Components of CARA such as OASYS [10] and T-REX [11] extract valuable data from open sources in real time.

The SOMA Terror Organization Portal (STOP) 2 SOMA Extraction Engine The SOMA Extraction Engine (SEE) is intended to derive SOMA-rules automatically from real-time sources such as T-REX [10] and OASYS[11]. These systems automatically extract several kinds of events. They are being tuned to extract social science codes such as those used in the MAROB effort. Till that is complete, the SOMA extraction engine is directly working with the MAROB data to extract rules.

A SOMA rule about a group G has the form Action:[L,U] if Env-Condition


− Action is an action that the group took (such as KIDNAP) − Env-Condition is a logical conjunction of elementary conditions on the environmental attributes. An elementary condition associated with the environmental attribute A is an expression of the form A op value where op is in the set { =, =, = }.

− [L,U] is a closed sub-interval of the [0,1] interval.

The above rule says that in any year when the Env-Condition is true, there is a probability between L and U that the group took the action stated in the rule. The rule below is an example of a rule that we extracted about Hezbollah.

KIDNAP: [0.51,0.55] if solicits-external-support & does not advocate democracy.

This rule says that in years when Hezbollah both solicited external support and did not promote democratic institutions, there was a 51 to 55% probability that they engaged in kidnapping as a strategy.

The SEE Algorithm works in accordance with the sketch given below. It takes as input, a group G, along with a table Tab(G), for the group. This table contains one column for each attribute of this group, and one row for each year. The attributes are split into action attributes (which describe the actions the group took in a given year) and environmental attributes (which describe the environment in which the group functioned during that year – these can include actions that other organizations took against group G). SEE takes an action name a and a code c for that action as input (e.g.

KIDNAP may be an action and the code 0 might indicate the code for no kidnappings). Based on these inputs, SEE attempts to find all conditions on the environmental attributes that predict KIDNAP=0 (i.e. that kidnappings will not be resorted to by the group) with a high degree of probability. Clearly, we are only interested in conditions which, when true, predict KIDNAP=0 with high probability, and which when false, predict KIDNAP=1 with high probability. In conditional probability terms, we are interested in finding conditions C on the environmental attributes such that abs(P(KIDNAP=0|C) – P(KIDNAP=0|~C)) is high. As usual, P(KIDNAP=0|C) denotes the conditional probability of the value of the KIDNAP attribute being 0 in a given year, given that condition C holds for that year.

Of course, even when we find such a condition C above, we would like to ensure adequate support, i.e. to ensure that there is a sufficiently large number of years in which both C and ~C are true.

The SEE system implements the algorithm in [2] to ensure that the conditions found satisfy both these requirements.

As of the time of writing this paper (early Dec. 2007), SEE has extracted behavioral models of 23 groups worldwide. The list includes 8 Kurdish groups spanning Iran, Turkey, and Iraq (including groups like the PKK and KDPI), 8 Lebanese groups (including Hezbollah), several groups in Afghanistan, as well as several other Middle Eastern groups.

3 SOMA Analyst NEtwork (SANE)

STOP requires an analyst to register and log in prior to accessing the system. Once the user is logged in, she proceeds via four phases.

Phase 1. Selecting a Group.

In this phase, the user selects a group to study by examining a drop down list of groups that have been processed so far. Figure 1 below shows this screen when the user has selected the PKK as his group of interest.

Figure 1: The STOP user has selected the PKK for analysis.

The SOMA Terror Organization Portal (STOP) Phase 2. Selecting an Action. Once the user selects a group, a list of actions that this group has engaged in during the period of study (typically the last 25 years or a bit less, depending on how long the group has been in existence) pops up. The user can select one or more actions to study. Figure 2 shows this screen when the user has selected the action “Arms trafficking.” Note that there is a number next to each action.

This number shows the number of rules we have for the group for this action. For instance, Figure 2 shows that we have 28 rules about “Armed Attacks” for the PKK.

Figure 1 The STOP user has selected the action Arms trafficking. There are 10,873 rules about "Arms trafficking" in the system.

Phase 3. Selecting Conditions.

Once the user selects an action(s), the system tries to find all rules extracted by SEE about the group dealing with the action(s) selected in Phase 2. Phase 3 allows the user to further focus on those SOMA rules that have certain antecedents. Figure 3 shows what happens when we continue looking at the PKK using the action “Arms trafficking” and we select the condition “Does not advocate wealth redistribution”. This selection basically says: Find all rules about the PKK that SEE has derived which deal with the action arms trafficking where one of the antecedents of the rule is does not advocate wealth redistribution. Once this query has been formulated and submitted, the system computes a web page about the PKK.

Figure 2 The STOP user has selected "Does not advocate wealth redistribution as the antecedent in their rules about Arms trafficking.

Figure 4 shows the result of the above inquiry by an analyst. The panel on Figure 4 is divided up into multiple parts.

Recent News. The “Recent News” panel on the result window shows recent news about the PKK. Recent news is obtained by accessing online news wires to present the analyst with the most recent news on an organization, irrespective of whether it deals with the query or not.

Extracted SOMA Rules. The “Extracted SOMA Rules” panel shows all the SOMA rules extracted for the analyst’s inquiry. Our default is to order these in descending order of probability (though other options are available to be discussed below). For instance, the first rule shown in Figure 4 tells us that there is an 88.23% probability

that the PKK engages in the arms trafficking action when:

− They do not advocate wealth redistribution and − They receive foreign state support and − There is no intra-organizational conflict.

The bottom right of the screen shows that there were a total of 82 SOMA rules we derived for this query and that the first 25 are shown on this screen. The user can browse all these 25 rules on this screen or move to another screen by transitioning to another page (see bottom left of the panel for how to move from one page to another).

The SOMA Terror Organization Portal (STOP) Basic Information about the Group. The bottom of the page also shows that STOP can access other sources such as Wikipedia in order to present the user some background information about the group. Sources other than Wikipedia are also planned for access in coming months Access to other Systems. Other systems such as T-REX [10] can provide valuable information about activities that different groups around the world might be Figure 3. The results from a query in the STOP system engaging in. For example, the T-REX system can automatically extract which violent activities groups are involved in. T-REX processes around 150K pages each day from approximately 145 news sources worldwide.

In addition, the CONVEX system [12] under development at the University of Maryland provides an alternative to SOMA for forecasting. In a future version of STOP, we anticipate providing access to CONVEX through the SOMA Terror Organization Portal. We also hope to provide access to a wide variety of third party systems as well, but this is not the case at this point.

SANE contains facilities for social networking amongst national security analysts and researchers working on counter-terrorism efforts worldwide. The kind of social networking supported by SANE is shown at the bottom on Figure 4.

− Mark as Useful. Analysts can mark certain rules as “Useful”. When an analyst marks certain rules as useful, a tuple gets stored in our SOMA Utility Database underlying STOP. This tuple identifies the group, the id of the rule, and the userid of the user who found the rule to be useful. In addition, the analyst may enter a comment about why he or she found the rule useful. This too is stored. Additional metadata about the transaction is also stored.

− Mark as Not Useful. Just as in the preceding case, analysts can mark rules as “Not Useful” as well.

− Show Ratings. SANE typically shows rules to an analyst in descending order of probability. However the design of SANE also allows a “Show Ratings” capability where the analyst can see ratings that other analysts have given to rules. We expect, in the very near future, to allow users to query the SOMA Utility Database so that he can see which rules have been marked as interesting or not interesting by colleagues’ whose opinion he respects.

The “Yearly Data” tab towards the bottom left of the screen shows yearly data (i.e. the actual Minorities at Risk Organizational Behavior codes for a given year). The SANE design allows an analyst to pull these up, mark them as useful or not, and also insert comments on the codes. These features, which are in the process of being incorporated into the system, will, upon completion, allow further social networking activities along the lines mentioned above.

4 SOMA Analyst Forecast Engine (SAFE)

Pages:   || 2 |

Similar works:

«Effect of Inulin on the Survival of Lactic Acid and Probiotic Bacteria in Ice Cream By Najah Boughida A Research Paper Submitted in Partial Fulfillment of the Requirements for the Master of Science Degree In Food & Nutritional Sciences Approved: 2 Semester Credits ~ Naveen Chikthimniah, Ph.D. The Graduate School University of Wisconsin-Stout May, 2011 The Graduate School University of Wisconsin-Stout Menomonie, WI Author: Najah Boughida Title: Effect of Inulin on the Survival of Lactic Acid and...»

«. A Newsletter of First Presbyterian Church of Ann Arbor October 2015 Volume 63, Number 8 WHAT’S INSIDE Worship & Music Annual Giving Faith in Action Opportunities for Women. 5 Adult Education Campus Connection Youth Connection Congregational Life Transitions October is a beautiful month in Ann Arbor. It’s the time when we settle into the rhythms of a new school year but do not yet have the joys and the pressures of the holidays. We begin the month with World Communion Sunday and a new...»

«2. Полимерные композиционные материалы: структура, свойства, технология: учеб. Пособие / [М. Л. Кербер, В. М. Виноградов, Г. С. Головкин и др.; под ред. А. А.Берлина]. – СПб.: Профессия, 2008. – 560 с., ил.3. Букетов А. В. Дослідження фізико-механічних властивостей при згині системи...»

«Adaptive curvelet-domain primary-multiple separation Felix J. Herrmann∗, Deli Wang† and Dirk J. (Eric) Verschuur‡ (January 18, 2008) Running head: Curvelet-matched filter ABSTRACT In many exploration areas, successful separation of primaries and multiples greatly determines the quality of seismic imaging. Despite major advances made by Surface-Related Multiple Elimination (SRME), amplitude errors in the predicted multiples remain a problem. When these errors vary for each type of...»

«Joint Barents Transport Plan Proposals for development of transport corridors for further studies September 2013 Front page photos: Kjetil Iversen, Rune N. Larsen and Sindre Skrede/NRK Table of Contents Summary 7 1 Introduction 12 1.1 Background 12 1.2 Objectives and members of the Expert Group 13 1.3 Mandate and tasks 14 1.4 Scope 14 1.5 Methodology 2 Transport objectives 15 2.1 National objectives 15 2.2 Expert Group’s objective 16 3 Key studies, work and projects of strategic importance 17...»

«Die 15 wichtigsten Methoden und Werkzeuge im Prozessmanagement Zusammenfassung Schnell und kostengünstig, aber dabei qualitativ hochwertig produzieren. So lautet das Motto der heutigen Zeit. Die Herausforderungen für Unternehmen sind gestiegen – immer größere Märkte bedeuten wachsende Konkurrenz, zudem werden die Anforderungen der Kunden individueller. Hier ist ein intelligentes Prozessmanagement gefragt, das die Fertigung so schlank, wie möglich, maximal flexibel und ressourcensparend...»

«Social Capital and Networks: A Proposal for Conceptual Integration Carsten Herrmann-Pillath, Witten/Herdecke University chepi@uni-wh.de Inhaltsverzeichnis Abstract Zusammenfassung 1 The Magic of Words and the Tower of Babylon 1.1 The rise of a transdisciplinary idea 1.2 A conceptual chameleon? 2 Defining social capital 2.1 Social capital as structural property of network agents 2.2 Social capital and power: a stock/flow approach 2.3 First consequences for the measurement of social capital 3...»

«User Manual for PLC Programming with CoDeSys 2.3 Copyright  1994, 1997, 1999, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2010 by 3S-Smart Software Solutions GmbH All rights reserved. We have gone to great lengths to ensure this documentation is correct and complete. However, since it is not possible to produce an absolutely error-free text, please feel free to send us your hints and suggestions for improving it. Trademark Intel is a registered trademark and 80286, 80386, 80486, Pentium are...»

«DISS. ETH NO. 21111 FRAGMENTATION GENETICS OF TROPICAL TREE SPECIES IN AN AGRO-FOREST LANDSCAPE A dissertation submitted to ETH ZURICH for the degree of Doctor of Sciences presented by SASCHA A. ISMAIL Dipl. Umwelt-Natw. ETH born 29th June 1979 citizen of Trub (Bern) accepted on the recommendation of Prof. Dr. Jaboury Ghazoul, examiner Dr. Chris J. Kettle, co-examiner Prof. Dr. Olivier J. Hardy, co-examiner Prof. Dr. Alex Widmer, co-examiner TABLE OF CONTENTS SUMMARY 1 ZUSAMMENFASSUNG 3 CHAPTER...»

«An Overview of Glaciers, Glacier Retreat, and Subsequent Impacts in Nepal, India and China WWF Nepal Program March, 2005 CORRECTION On page 29 of the following report WWF included the following statement: In 1999, a report by the Working Group on Himalayan Glaciology (WGHG) of the International Commission for Snow and Ice (ICSI) stated: `glaciers in the Himalayas are receding faster than in any other part of the world and, if the present rate continues, the livelihood[sic] of them disappearing...»

«GEOLOGICAL SURVEY OF IRELAND (GSI) ANNUAL REPORT 2006 APPENDICES This information supplements that in the published GSI Annual Report for 2006, copies of which can be obtained from the GSI Customer Centre. Additional information on the work of GSI is available through “Geology Matters”, the regular GSI Newsletter, copies of which can be accessed through www.gsi.ie or by contacting the GSI Customer Centre, Geological Survey of Ireland, Beggars Bush, Haddington Road, Dublin 4. Phone (01)...»

«THIS CIRCULAR IS IMPORTANT AND REQUIRES YOUR IMMEDIATE ATTENTION If you are in any doubt as to any aspect of this circular or as to the action to be taken, you should consult your licensed securities dealer, bank manager, solicitor, professional accountant or other professional adviser. If you have sold or transferred all your shares in China Innovationpay Group Limited (the “Company”), you should at once hand this circular, together with the enclosed form of proxy, to the purchaser or...»

<<  HOME   |    CONTACTS
2016 www.abstract.xlibx.info - Free e-library - Abstract, dissertation, book

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.