Title: Comparing trends in topics in economic journals overtime
Authors: Peter Winker - University of Giessen (Germany) [presenting]
David Lenz - Justus-Liebig University Giessen (Germany)
Abstract: The comparison of information content of different sources is highly relevant. We compare text corpora, specifically articles published in two economic journals. Thereby, the focus is on the development of topic importance over time and how it correlates across journals. Similar questions arise in many fields of practical relevance and, if at all, are mostly answered impressionistically. We present a quantitative framework for comparing text corpora based on their latent topics using text mining techniques. Paragraph Vector Topic Modeling is applied to identify latent topics in text corpora and time information is utilized to track the evolution of these topics. This allows the comparison of corpus compositions over time. More specifically, we evaluate three comparison methods: Treat both text corpora as a single corpus, train a model on one corpus and evaluate the other corpus based on this model and vice versa, and train a model for each corpus and use a matching approach for pairing corresponding topics. For the empirical application, we exploit the corpus of articles published in the Journal of Economics and Statistics and the corpus of articles published in the Review of World Economics, both from 1913 to 1940. We present topic dynamics for both corpora and information on how strong the correlation of these dynamics was across journals. Furthermore, the analysis indicates which of the methods presented above is most promising for this type of analysis.