Title: Analyzing economic texts using network based topic models
Authors: Ryohei Hisano - University of Tokyo (Japan) [presenting]
Abstract: Topic models are one of the most commonly used statistical modelling approaches when analyzing economic and financial texts, due to its high interpretability and efficient nature. However, when analyzing economic texts, basic information besides the good old bag-of-words matrix, that topic models aim to model, becomes important. These additional information includes information such as who mentioned what at what timing, what were the economic indicators when a text was written, or even simply the meaning of a word to just name a few. We show a simple network based approach to incorporate these additional information into a topic model and observe what additional insights could be gained from our approach. We would mainly focus on analyzing the Economy watcher survey, published by the Cabinet office of Japan, but other data sets might be mentioned if time permits.