Jafar I. Adibi

 
   
International Relation
  
  Lectures / Talks  
 
  • Middle East Institute
  • UCLA please contact me for lecture slides
  
  Resersach Interests  
 

I am interested in application of data mining and knowledge discovery in Conflict Resolution, Early Warning Systems and Security for International Conflicts.

I have been active in two major projects with the Center for International Studies (CIS) at USC and Annenberg School for Communication :

  • Paris in Los Angeles (PARIS)
  • Conflict Early Warning System) CEWS
 
 

Introduction to Early Warning through Data Mining

 
 
How should the international community respond to the next Somalia disaster, the next Bosnia, the next Haitian crisis? In the post-Cold War era, these "new security agenda" questions are especially troubling, because Cold War Great Power conflicts have little to tell us about what should be done, and by whom. Yet considerable relevant information on these kinds of conflicts already exists, much of it in computer-readable formats; and the experience of international agencies in coping with such issues goes back in many cases to the pre-Cold War period. Using and improving upon such information, as well as the latest developments in computerized informational handling software engineering, we propose to develop a two-tier information system oriented toward exploring and recommending actions to be taken by international agencies responsible for international or transnational conflict management and prevention, such as the United Nations Security Council or the United Nations High Commissioner for Refugees. These are two of the inter-governmental agencies with the long track records in coping with such cases, and much of what they have learned could be of value for other prevention or management oriented nongovernmental agencies as well.

Much of the reasoning in practical conflict research is based on precedential analysis, examining previous situations, drawing parallels, and adapting past solutions to current problems. In Artificial Intelligence (AI) and Data Mining , case-based reasoning research has explored this style of problem solving. We propose to develop an AI system that will assist mediators, conflict managers and preventers find alternative solutions and hypothetically explore their potential consequences on the basis of historical precedents.

We propose to build an extensible prototype system with two interconnected parts:

1. A case-based reasoning system for computer-assisted explorations of several dozen past conflict cases, pointing toward syntheses of action recommendations and early warning system for newer, emerging or hypothetical target situations;

2. A Notebook hypertext information base containing larger textual and encoded information bases of potentially relevant past experience, as available from disclosable databases and on-line Internet information sites.

The interface connecting the two systems would allow selective transfer of new cases from the hypertext information component into the AI case-based reasoning subsystem for more intensive exploration purposes, although new codification of such information might be required. The prototype will address many challenging problems that are currently on the cutting edge of AI and hypertext systems research, including reasoning from multiple cases, learning and discovery, integrating sub-cases, exploring alternative historical trajectories, and the reorganization of case histories, text mining and discovering knowledge form semi-structured databases such as Internet.

Our effort intends to build on the recent acceptance of the use of advanced computer and communication technologies by international organizations in the conflict resolution area. Stanford Unviersity has an excellent back ground on Data Mining and International Relation fields. We feel the inter-disciplinary competence we bring to this study will produce a significant and useful technological product in this urgent area of concern to the international community.

Knowledge Discovery and Data Mining

With the emphasis on collecting data increasing around the world, there is an urgent need for a new generation of different techniques, methods and algorithms to assist researchers, analysts, decision makers and managers in extracting useful patterns from the rapidly growing volumes of data. These techniques and tools are the subject of the emerging field of knowledge discovery in databases (KDD). KDD has evolved from interaction and cooperation among such different fields as machine learning, pattern recognition, database, statistics, artificial Intelligence, knowledge representation, and knowledge acquisition for intelligent systems. The main idea in KDD is to discover a high level knowledge (abstract knowledge) from lower levels of relatively raw data, or to discover a higher level of interpretation and abstraction than those previously known.

During the past 5 decades the notion of finding or discovering useful interesting patterns in data has been addressed by different research groups and researchers; we wish here to give a better idea of how KDD relates to these other approaches. Such approaches have been given different names, such as exploratory data analysis, information discovery, information harvesting, data archaeology, and data pattern recognition. KDD applies machine learning and pattern recognition techniques to extract patterns implicit in a database. The new wave of KDD addresses the overall process of discovering useful knowledge from data while data mining, statistic analysis and other such techniques address only a particular step in this process. KDD seeks incrementally to understand, to adapt and apply these patterns to future cases or data sets. KDD uses statistical methods, especially exploratory data analysis methods, but it sees their use as only one part of a more comprehensive knowledge discovery process. Statistical methods and algorithms offer precise methods for quantifying inherent inferential uncertainties. KDD systems embed particular statistical procedure for and modeling data, evaluating hypotheses and handling noise within an overall knowledge discovery framework. KDD approaches and methods are focussed on model extraction or construction or discovery, rather than the parameter estimation of previously hypothesized models. They operate best in the context of large sets with rich data structures. For such large data sets, interpretations may already exist, coming from a special field of inquiry; by shifting the window of concern to another aspect of that data base, we may get some new pattern for another purpose.

There are several knowledge discovery and analysis issues and questions which would like to address. Some of these were directly related to the main purpose of our efforts; others arose in the course of the project. These include:

  • · How could we describe conflicts in terms of features or attributes in the most appropriate way?
  • Given a data base of well described cases, we are often presented with a new case. Which prior case is most similar to the new case?
  • How could we define a good overall distance or proximity measure for conflicts or management/prevention packages?
  • Can this measurement approach be extended to more specifically characterized aspects or features of especially related conflicts or conflict management processes?
  • How can we compare the different modalities of textually describing, codifying or classifying conflicts?
  • After all these issues of comparison and proximity measurement are resolved, we are still faced with classification and clustering issues: How can we most accurately and usefully classify conflicts and cases?
  • Can we detect patterns associating conflict attributes and conflict-tools employed by conflict managers?
  • How can we proceed when recodifying a large set conflicts in terms of someone else’s conceptual scheme is so terribly costly and time consuming?

To address the above mentioned questions we would like to classify and cluster international conflicts and conflict prevention efforts on the basis of information assembled in different databases. It has sought to apply versions of case-based reasoning techniques to suggest ways of decreasing the level of violence in emergent or ongoing violent conflicts. In contrast to the other preliminary explorations of the utility of knowledge discovery techniques like decision trees, we have not be interested in merely describing different patterns in our data such as "What is the pattern of participation of US, USSR and China in Third World conflicts? " we have a more specific and heuristic conflict prevention focus: we will try to gear our analyses to preventively oriented concerns, to the discovery of actions which conflict managers or preventers have taken, or could have taken, to prevent violence escalation and encourage violence diminution. In addition we try to provide an early warning system for conflict prevention through text mining over web data. Our exploration and development of KDD techniques is defined through such lenses.

In contrast with traditional data analysis, the KDD process is interactive and iterative. One has to make several decisions in the process of KDD. In the following we explain the major steps in KDD.

Understanding the domain knowledge and identifying the goal of KDD: in this proposal the main emphasis is on the analysis of the international conflicts and on the use of management and violence prevention tools by international management agencies.

  • Creating the main data set for the purpose of the KDD: in this project we propose to use of several source of data bases: such as SHERFACS, CASCON and HAAS.
  • Finding interesting features in the database. This step is extremely important in the field of International Studies. Researchers and practitioners with different backgrounds and different languages may work on a given database, getting different results. Each group may consider different attributes in doing so. · Apply a data-mining algorithm such as clustering, classification, regression, in ways which match with the original goal of the KDD process.
  • Data mining and searching for interested patterns in the data.
  • Interpretation of the result or found pattern. A good clustering or classifying approach should explain the result of such an approach. If a result can not be interpreted properly, it may not be useful for further purposes.
  • Prepare a new set of knowledge for future analysis, utilization or discovery purposes.