Bayesian Changepoint Model for Text Data

Posted: 31 May 2019

See all articles by Tomoya Sasaki

Tomoya Sasaki

Massachusetts Institute of Technology (MIT)

Daichi Mochihashi

The Institute of Statistical Mathematics

Date Written: May 23, 2019

Abstract

Exploring and identifying structural changes in the characteristics of or relationships between political actors is one of the fundamental goals of political science. Some changes can be identified using documents (latent traits) include positions of parties/legislators from manifestos/press releases, and relationships between countries from treaties. The proposed model simultaneously estimates topics (contents of the documents), the location of changepoints, and the number of changepoints (no need to compare multiple models). It considers multiple latent traits in documents (a topic-model based approach), shows which topics (characteristics) contribute to change, and identifies multi-level changes (hierarchical changepoint).

Keywords: bayesian, changepoint, model, data

Suggested Citation

Sasaki, Tomoya and Mochihashi, Daichi, Bayesian Changepoint Model for Text Data (May 23, 2019). MIT Political Science Department Research Paper No. 2019-16, Available at SSRN: https://ssrn.com/abstract=3393076 or http://dx.doi.org/10.2139/ssrn.3393076

Tomoya Sasaki (Contact Author)

Massachusetts Institute of Technology (MIT) ( email )

77 Massachusetts Avenue
50 Memorial Drive
Cambridge, MA 02139-4307
United States

Daichi Mochihashi

The Institute of Statistical Mathematics ( email )

4-6-7 Minami-Azabu
Minato-ku, Tokyo 106
Japan

Do you have negative results from your research you’d like to share?

Paper statistics

Abstract Views
621
PlumX Metrics