| 1 |
Can we forecast conflict? A framework for forecasting global human societal behavior using latent narrative indicators
|
|
|
|
Abstract:
The ability to successfully forecast impending societal unrest, from riots and protests to assassinations and coups, would fundamentally transform the ability of nations to proactively address instability around the world, intervening before unrest accelerates to conflict or prepositioning assets to enhance preventive activity. It would also enhance the ability of social scientists to quantitatively study the underpinnings of how and why grievances transition from agitated individuals to population-scale physical unrest. Recognizing this potential, the US government has funded research on “conflict early warning” and conflict forecasting for more than 40 years and current unclassified approaches incorporate nearly every imaginable type of data from telephone call records to traffic signals, tribal and cultural linkages to satellite imagery. Yet, current approaches have yielded poor outcomes: one recent study showed that the top models of civil war onset miss 90% of the cases they supposedly explain. At the same time, emerging work in the economics disciplines is finding that new approaches, especially those based on latent linguistic indicators, can offer significant predictive power of future physical behavior. The information environment around us records not just factual information, but also a rich array of cultural and contextual influences that offer a window into national consciousness. A growing body of literature has shown that measuring the linguistic dimensions of this real–time consciousness can accurately forecast many broad social behaviors, ranging from box office sales to the stock market itself. In fact, the United States intelligence community believes so strongly in the ability of surface-level indicators to forecast future physical unrest more successfully than current approaches, it now has an entire program devoted to such “Open Source Indicators.” Yet, few studies have explored the application of these methods to the forecasting of non-economic human societal behavior and have primarily focused on large-bore events such as militarized disputes, epidemics, and regime change. One of the reasons for this is the lack of high-resolution cross-national longitudinal data on societal conflict equivalent to the daily indicators available in economics research. This dissertation therefore presents a novel framework for evaluating these new classes of latent-based forecasting measures on high-resolution geographically-enriched quantitative databases of human behavior. To demonstrate this framework, an archive of 4.7 million news articles totaling 1.3 billion words, consisting of the entirety of international news coverage from Agence France Presse, the Associated Press, and Xinhua over the last 30 years, is used to construct a database of more than 29 million global events in over 300 categories using the TABARI coding system and CAMEO event taxonomy, resulting the largest event database created in the academic literature. The framework is then applied to examine the hypothesis of latent forecasting as a classification problem, demonstrating the ability of a simple example-based classifier to not only return potentially actionable forecasts from latent discourse indicators, but to quantitatively model the topical traces of the metanarratives that underlie them. The results of this dissertation demonstrate that this new framework provides a powerful new evaluative environment for exploring the emerging class of latent indicators and modeling approaches and that even rudimentary classification-based models may have significant forecasting potential. ; U of I Only ; Author requested U of Illinois access only (OA after 2yrs) in Vireo ETD system
|
|
Keyword:
Big data; Content analysis; Event analysis; Political forecasting; Text classification
|
|
URL: http://hdl.handle.net/2142/95525
|
|
BASE
|
|
Hide details
|
|
|
| 4 |
Workset Creation for Scholarly Analysis and Data Capsules (WCSA+DC): Laying the foundations for secure computation with copyrighted data in the HathiTrust Research Center, Phase I
|
|
|
|
BASE
|
|
Show details
|
|
|
| 7 |
Selected papers from Digital Humanities 2007, University of Illinois, Urbana-Champaign, 2 - 8 June 2007
|
|
|
|
BLLDB
|
|
Show details
|
|
|
| 9 |
Digital Humanities 2007 Conference Abstracts, Second Edition
|
|
|
|
BASE
|
|
Show details
|
|
|
|
|
