Short Term Prediction of International Events Using Pattern Recognition

使用模式识别对国际事件进行短期预测

• 批准号: 8910738
• 负责人: Philip Schrodt
• 金额: --
• 依托单位: University of Kansas Main Campus
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1989
• 资助国家: 美国
• 起止时间: 1989-07-01 至 1991-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8910738&HistoricalAwards=false
• 关键词: Short; Term; Prediction; International; Events

This project focuses on the development of artificial intelligence methods for finding short-term patterns in international behavior. Much of the contemporary study of international politics focuses on "events": discrete interactions between nation-states. Events may be as dramatic as the outbreak of war or the signing of a major trade agreement, or as simple as one government congratulating another on its independence day. Individual events can be group into event sequences which describe more complicated behaviors such as conflicts, negotiations or changes in relations. Human observers of international behavior such as journalists, historians and political scientists know that international events follow patterns. A war will almost inevitably be preceded by escalating tensions; a trade agreement will be preceded by months of negotiations. These patterns provide enough regularity in international affairs that expert analysts usually have a good idea of what the likely next events will be, and can also detect unusual changes in behavior. This project will use pattern recognition methods developed in artificial intelligence to detect short-term regularities in contemporary events in the Middle East. The events will be coded from the Foreign Broadcast Information Service (FBIS) reports available from the U.S. Government Printing Office. FBIS reports several hundred events per day: these will be coded into standard categories of events, then the computer programs will look for repeated patterns in those events data. Those patterns can then be used to make predictions much as human analysts use the observed regularities in international behavior to make predictions. The project uses two artificial intelligence methods. One method constructs "partially-ordered event sequences" - - sequences of events which are observed repeatedly because certain events must be preceded by other events (for example, one must have negotiations before one can have a trade agreement). The other method uses "genetic algorithms", which assemble complex event sequences out of simpler sequences using a process resembling evolution. The Principal Investigator has already published several papers employing these methods on historical data; this project will be the first to use them on contemporary data. The project will make three contributions to the understanding of international behavior. First, it will be one of the first efforts to create a computer model of the day-to-day behavior of a contemporary international system, the Middle East. While international relations theory has provided a number of computer models for long-term trends in international politics, relatively little work has been done on short- term behavior. Second, the project will create general tools for finding patterns in political behavior. While these techniques are being developed to study international behavior, the same methods could be used to study any other political, economic or social behavior which can be described using events. Finally, the project will provide some new tools for dealing with large amounts of data generated on a daily basis using relatively inexpensive microcomputer equipment.

该项目的重点是开发人工智能方法，以发现国际行为的短期模式。当代国际政治研究的大部分都集中在“事件”上：民族国家之间的离散互动。事件可能像战争的爆发或重大贸易协定的签署那样戏剧性，也可能像一个政府祝贺另一个政府独立日那样简单。单个事件可以被分组为事件序列，描述更复杂的行为，如冲突，谈判或关系的变化。观察国际行为的人，如记者、历史学家和政治学家，都知道国际事件是有规律可循的。战争爆发前，紧张局势几乎不可避免地会升级;贸易协定签署前，将进行数月的谈判。这些模式在国际事务中提供了足够的规律性，专家分析师通常对接下来可能发生的事件有很好的了解，并且还可以发现行为的异常变化。该项目将使用人工智能中开发的模式识别方法来检测中东当代事件中的短期波动。这些事件将从美国政府印刷局提供的外国广播信息服务（FBIS）报告中编码。FBIS每天报告数百起事件：这些事件将被编码为标准的事件类别，然后计算机程序将在这些事件数据中寻找重复的模式。然后，这些模式可以用来做出预测，就像人类分析师使用观察到的国际行为中的行为来做出预测一样。该项目使用两种人工智能方法。一种方法构造“部分有序事件序列”-重复观察的事件序列，因为某些事件必须在其他事件之前发生（例如，在达成贸易协定之前必须进行谈判）。另一种方法是使用“遗传算法”，它使用类似进化的过程从简单的序列中组合出复杂的事件序列。主要研究者已经发表了几篇论文，采用这些方法对历史数据进行分析;本项目将是第一个将它们用于当代数据的项目。该项目将为理解国际行为做出三项贡献。首先，这将是创建当代国际体系（中东）日常行为的计算机模型的首批努力之一。虽然国际关系理论为国际政治的长期趋势提供了一些计算机模型，但对短期行为的研究相对较少。第二，该项目将创建用于发现政治行为模式的通用工具。虽然这些技术正在开发用于研究国际行为，但同样的方法可以用于研究任何其他可以使用事件描述的政治，经济或社会行为。最后，该项目将提供一些新的工具，用于处理使用相对廉价的微型计算机设备每天产生的大量数据。