Modeling and Forecasting Elections Using Topic Models
Keywords
Loading...
Authors
Issue Date
2014-07-08
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
After elections, people want to know who will win the election as quickly as possible.
During election night forecasts are made, based on both polls and the results from early
polling stations. In this research a forecasting model based on topic models is proposed.
The model's forecasting performance is compared to a linear model's performance using the
Dutch House of Representatives election. The model is also used to visualize and analyze
voting profiles. The proposed model outperforms existing linear models with a lower mean
absolute error if 10% or more of the polling stations are observed. For 2.5% or less observed
polling stations the linear model has a lower mean absolute error. The proposed model is
also able to give insight into voting behavior by visualizing voter pro les. Thus, the proposed
model is useful for both forecasting and modeling elections.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen