contributor:
Presented here are experiments on Brandes's Main Currents ... using structured topic models. Structured topic models, an approach pioneered by Margaret Roberts (UCSD) and Edo Airoldi (Temple University), allows the user to include confounding factors in the creation of topics models. Consequently, models can be explored not only as a simple representation of a standard LDA topic model, but also in the context of one or more confounding factors, such as time, gender of author, or some sort of affiliation of various classes of writers or sources.
Several browsers based on STM are presented here. The first two browsers use the mention of Goethe in the Hovedstrømninger chunks as a confounding factor. The remaining three browsers allow for the exploration of a series of additional models, with k=84 in browser three
All of these browsers rely on the STMBrowser package developed by Roberts (https://github.com/mroberts/stmBrowser). The goal of these approaches are to understand how topics vary over various confounding factors, be those factors based on time, gender, age, affiliations, or any other possible attribute or feature. Additional experiments using STM will be posted as our work develops.
Structured TM Browser: www.structuraltopicmodel.com
Authors: Michael Freeman, Jason Chuang, Molly Roberts, Brandon Stewart & Dustin Tingley
Please email all comments/questions to molly.e.roberts [at] gmail.com
© Det Danske Sprog- og Litteraturselskab med støtte fra Carlsbergfondet • Privatlivspolitik
