Automatic thematic classification of election manifestos
作者：
Highlights：
• We digitized political texts from the 1980s and 1990s.
• We used these data to learn a classifier that can label more recent political texts.
• Change of themes over the years affects recall of the learned classifier.
• But precision is comparable to the precision obtained by a human expert labeller.
• For political themes, a high level of detail seems to be preferred by domain experts.
摘要
•We digitized political texts from the 1980s and 1990s.•We used these data to learn a classifier that can label more recent political texts.•Change of themes over the years affects recall of the learned classifier.•But precision is comparable to the precision obtained by a human expert labeller.•For political themes, a high level of detail seems to be preferred by domain experts.
论文关键词：Text classification,Political data,Expert evaluation
论文评审过程：Received 27 March 2013, Revised 30 August 2013, Accepted 26 February 2014, Available online 13 April 2014.
论文官网地址：https://doi.org/10.1016/j.ipm.2014.02.006
