摘要
大数据的出现给国际政治预测带来了新的希望,但大数据并非无所不能。其成功预测的前提是事件具备起码的稳定性和连续性。作为一个过程,大数据预测大致包括数据准备、分析建模和模型应用与反馈三个阶段。数据准备主要由数据获取和数据预处理两个环节组成。对于数据获取而言,研究者既面临着出于保护个人隐私和国家安全的需要所施加的规范约束,又要努力克服数据资源的结构性缺陷所造成的现实约束。数据预处理则涉及通过数据挖掘技术从原始数据中提取特定数据的特征工程和旨在提高信噪比的数据降噪。在分析建模阶段,研究者设置的算法和模型会显著地影响到预测的效果。在模型应用和反馈阶段,研究者首先使用模型进行预测,然后根据预测结果来检验、评估和调试模型,其中事件背景条件和对象运行轨迹的变化是影响预测准确度的重要因素。从经验上看,上述诸条件满足得越多,预测准确率越高。本文按照大数据预测的工作流程,归纳并分析了国际政治事件预测实践中各个环节所面临的约束条件。文章创新之处在于较为深入地探讨了因果关系在大数据预测中的作用:它不仅是建模的基础,而且深刻地影响到预测的整个过程。
Big data has become a new instrument for international political forecasting. The success in big data forecasting is premised upon the minimally acceptable stability and continuity of an event. Big data forecasting is made up of three stages,namely data preparation,model building and model application. Data preparation is composed mainly of data access and pretreatment,at which stage the researchers are confronted with both normative and substantive constraints. At the stage of model building,the researchers construct the algorithms and models,which would significantly influence the effectiveness of forecasting. At the stage of model application and feedback,the researchers would first make predictions on the basis of their models,and then test,evaluate and adjust the modes in terms of the results. Among other things,the contextual information of an event and the path of an object play a substantial role in accurate prediction. Empirically speaking,the more the above constraints have been overcome,the more accurate a forecasting would be. Additionally,big data forecasting is evaluated predominantly by cross-validation,which does not require a deep exploration into the connections between variables. Moreover,causality is not always indispensable for the prediction of some events. Nonetheless,causality is merely the foundation of model building,but would exert profound effects upon big data forecasting during the whole process.
作者
卢凌宇
张传基
LU Lingyu;ZHANG Chuanji
出处
《欧洲研究》
CSSCI
北大核心
2021年第4期130-154,I0005,共26页
Chinese Journal of European Studies
