摘要
通过机器学习来预测冲突正在成为当前国际关系研究领域的一个热议话题。但是从跨学科交叉研究的视角来看,计算机介入政治分析和国际关系研究并不是一个新现象,其间经历了从计算机模拟冲突场景到机器学习自动识别冲突模式的复杂变革历程。二者的共同点是都重视仿真社会互动情景与政治复杂演进过程,反对有关政治冲突现象的简单线性解释;但二者在研究取向上还是有着本质的不同,计算机模拟提倡基于特定社会理论的情景建模与逻辑推演,而机器学习则强调无特定社会理论支撑的特征识别与关联预测。有鉴于此,本文首先分析了计算机模拟与机器学习在冲突预测中的研究路径差异,然后重点阐述了无理论支撑下将机器学习应用于冲突预测之可能,并以2010—2016年印度恐怖袭击预测为例,实证检验了基于BP神经网络的机器学习在真实社会情景中的实际冲突预测效力,结果发现基于机器学习的冲突预测范式即使在没有特定社会理论支撑下,也具备一定冲突预测能力,并可产生新的冲突知识发现。但即便如此,作为一种跨学科交叉研究范式,机器学习介入冲突预测仍然面临重重困难。
Can machine learning predict armed conflicts?Or,can machine learning bring more accuracy in conflict forecasting?It is an emerging and fascinating topic for IR scholars.But,for IR's cross disciplinary research,it is a long journey from computer simulation on conflict episodes to machine learning of conflict recognition.Although both machine learning and computer simulation emphasize modeling the social reality and political process and oppose simple and liner explanations on conflicts,they represent different paradigms:the computer simulation needs to model on the basis of specific social theories,but the machine learning needs nothing to predict.Based on the above considerations,this paper firstly reviewed the academic process from computer simulation to machine learning in political analyses and IR studies,and compared the methodology change caused by machine learning and computer simulation;then,this paper elucidated the possibility and principle of conflict prediction based on machine learning.In the section of case study,this paper examined the efficiency of machine learning approach by using one-layer BP networks to predict India's terrorist attack from 2010 to 2016.The result shows that machine learning can predict the security situation.Finally,it concludes that it is possible to predict conflicts and produce new knowledge via machine learning,although it still faces various difficulties.
出处
《世界经济与政治》
CSSCI
北大核心
2017年第7期100-117,共18页
World Economics and Politics
基金
国家社科基金青年项目“基于大数据驱动的外交决策模式创新与我国实践路径研究”(项目编号:15CGJ034)
对外经济贸易大学中央高校基本科研业务费专项资金资助(批准号:CXTD8-05)之阶段性成果
关键词
冲突预测
机器学习
神经网络
恐怖袭击
conflict prediction
machine learning
neural networks
terrorist attack
