Topic Modeling of Popular Science Magazine Issues with Latent Dirichlet Allocation (LDA)

Abstract

TÜBİTAK is an institution that pioneers science in Turkey. In addition, it offers the opportunity to learn science at their level for various age groups with the journals it publishes. The scientific topics covered in the journals published by TÜBİTAK have evolved over the years, reflecting the dynamic nature of science. When examining literature studies on popular science, numerous studies are encountered. Popular science books and magazines are generally reviewed in these studies according to various subjects. However, studies that examine popular science magazines with the machine learning method are limited. It is essential to reveal how continuous and high-quality journals shape popular science in the context of projecting the popularity of future scientific topics. The Latent Dirichlet Allocation algorithm has been used intensively in topic modeling for many years. Within the scope of this study, multiple topics were identified for each journal by applying Latent Dirichlet Allocation to the textual content of the journals, and the distribution of each topic within the journal was calculated. As a result, this study automatically and independently extracted the topics and themes in the journals published by TÜBİTAK. These results allowed us to gain insight into how various content areas have evolved throughout the journal's history. With the results obtained from this study, new researchers can use the data to examine how topic change and/or topic expansion occur in a field.