非監督與監督式學習法訓練民主指標

Combination of Supervised and Unsupervised Learning for Training the Democracy Index

林天中(Tian-Chung Lin)

指導教授 ： 童涵浦

共同指導教授 ： 馮勃翰

國立臺灣大學/社會科學院/政治學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902146

摘要

民主指標為所有量化民主研究之基礎，如何建立一個客觀反映民主相關資料結構的民主指標為本文探討之問題。本文根據Gründler與Krieger提出之觀點，將資料透過機器學習的演算法導出民主指標，可以避免民主指標在加總計算時，由研究者主觀判斷各要素指標之間的相對重要性，因而使民主指標無法客觀反映民主相關資料。然而，該文透過其他廣泛使用之民主指標中極端民主與極端非民主之國家年代作為預視資料，使預視資料仍隱含其他指標之主觀判斷之結果。故本文目的為透過非監督式學習的叢集分析建立預視資料，使預視資料更能反映資料本身的分布特性，找出民主相關資料中分布最極端之數據，再透過監督式學習的支援向量機和支援向量迴歸建立二分民主指標模型與連續民主指標模型。

關鍵字

民主指標 ； 非監督式學習 ； 監督式學習 ； 支援向量機 ； 叢集分析

並列摘要

Democracy index is the foundation of quantitative democracy research. How to build a democracy index which can objectively reflect the structure and feature of democracy data is the main question of this research. This paper will apply the concept of building democracy index by machine learning algorithm which is introduced by Gründler and Krieger. In their point of view, machine learning can avoid the problem of deriving an aggregation method subjectively. However, they do not apply the machine learning algorithm to the methodology of selecting priming data. Instead, they select extreme democracy and non-democracy country-year in other notable democracy index as their priming data, which may still cause the objective problem which they want to solve. In this paper, I will apply unsupervised learning algorithm to construct priming data, and apply the data into supervised learning algorithm to build a new democracy index which can subjectively summarize the democracy data.

並列關鍵字

democracy index ； unsupervised learning ； supervised learning ； support vector machine ； cluster analysis

參考文獻

Abu-Mostafa, Yaser S., Malik Magdon-Ismail, and Hsuan-Tien Lin, 2012, Learning from Data: a Short Course. AMLbook.com.

Acemoglu, Daron. , Suresh Naidu, Pascual Restrepo, and James A. Robinson. 2019. “Democracy Does Cause Growth.” Journal of Political Economy. 127(1): 47-100.

Anckar, Carsten, Cecilia Fredriksson. 2019. “Classifying political regimes 1800-2016: a typology and a new dataset.” European Political Science. 18(1): 84-96.

Boix, Carles, Michael Miller, Sebastian Rosato. 2013. “A Complete Data Set of Political Regimes, 1800-2007.” Comparative Political Studies. 46(12):1523-1554.

Brock, Guy, Vasyl Pihur, Susmita Datta, Somnath Datta. 2008. “clValid: An R Package for Cluster Validation.” Journal of Statistical Software. 25(4):1-22.