Using Natural Language Processing to Analyze Political Party Manifestos from New Zealand †

Abstract

This study explores how natural language processing (NLP) can supplement content analyses of political documents, particularly the manifestos of political parties. NLP is particularly useful for tasks such as: estimating the similarity between documents, identifying the topics discussed in documents (topic modeling), and sentiment analysis. This study applies each of these techniques to the study of political party manifestos. Document similarity may be used to gain some insight into the way parties change over time and which political parties are successful at bringing attention to their policy agenda. Categorizing text into topics may help objectively categorize and visualize the ideas political parties are discussing. Finally, sentiment analysis has the potential to show each political party’s attitude towards a policy area/topic. This study specifically applies these techniques to the manifestos produced by the political parties of New Zealand, from 1987 to 2017 (a period of significant party system change in New Zealand). It finds that NLP techniques provide valuable insights, although there is a need for significant fine-tuning.