| 曹玮,刘倩,刘子夜.朝鲜核行为预测：变化无常还是有章可循？[J].国际安全研究,2019,37(1):60-88 |
| 朝鲜核行为预测：变化无常还是有章可循？ |
| Capricious or Rule-based? Naive Bayesian Model and the Prediction of North Korea’s Nuclear Behavior |
| 修订日期：2018-12-02 |
| DOI：10.14093/j.cnki.cn10-1132/d.2019.01.003 |
| 中文关键词: 朝核问题 核试验 贝叶斯方法 预测方法 对外政策研究 |
| 英文关键词: North Korea’s nuclear issue, nuclear test, Bayesian method, prediction approach, foreign policy research |
| 基金项目: |
|
| 摘要点击次数: 1561 |
| 全文下载次数: 1055 |
| 中文摘要: |
| 学界普遍默认朝鲜对外政策难以预测，因此大量研究始终停留在“事后”诠释的水平，缺乏预判性、前瞻性研究。作者挑战了这一成见，建立了关于朝鲜核试验和导弹试验行为的短期预测模型。通过以2006～2018年朝鲜媒体国际新闻报道为数据集，以每月的数据作为一个数据样本，通过构建最优特征集，应用朴素贝叶斯方法，建立了朝鲜核行为的预测模型。经过测试，模型对朝鲜历史行为的“预测”整体准确率超过80%。预测效果良好，且模型具有稳健性。根据统计结果建立的预测朝鲜核行为的预警指标体系，可以从长期、中期和短期三个时段实时监测朝鲜核动向。为解决报道搜集相对滞后的问题，运用SARIMA时间序列分析方法，模拟补全出未来六个月的特征集数值，估算数据在统计上具备可信性，预测准确率高。研究发现，朝鲜核行为远非外界印象中那样无章可循，事实上，朝鲜在相当长时间段内的核行为呈现出高度稳定的模式化倾向。即使像朝鲜这样信息极度封闭的国家，采用适当方法仍然可以对其行为做出较准确的预测。这种建模方法能够为发展关于国家行为预测的一般方法提供启发。 |
| 英文摘要: |
| It is generally believed in academia that North Korea’s foreign policy is difficult to predict. Accordingly, a significant amount of research on its foreign policy has been subject to ex post facto interpretations without any predictability. This paper intends to challenge this preconception and thus establish a short-term prediction model for North Korea’s nuclear and missile tests. By using international news reports from North Korean media from 2006 to 2018 as a dataset and monthly data as the data sample and by creating an optimal feature set and applying the naive Bayesian method, a prediction model for North Korea’s nuclear behavior is able to be constructed. The model’s prediction tests show that the overall accuracy exceeds 80%. With the desirable prediction effects, the model is proved to have robust performance. Based on the statistical results, an early-warning index system for predicting North Korea’s nuclear behavior can be established to conduct real-time monitoring on North Korea’s nuclear trends on a short-term, medium-term or long-term basis. In order to solve the problem of relative lag in the collection of news reports, this paper attempts to adopt the SARIMA time series analysis to simulate the feature set data of the next six months. The estimated data are statistically credible and the prediction accuracy is relatively high. The findings are that North Korea’s nuclear behavior is far from being unpredictable as perceived by the outside. In fact, North Korea’s nuclear behavior exhibits a highly stable tendency of patternization over a fairly long period. This paper tries to prove that when appropriate methods are adopted, even for such information-closed countries as DPRK, relatively accurate predictions of state behavior will work out. This modeling approach can shed light on developing general approaches to the predictions of state behavior. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|