Temporary Topic Models in Social Sciences: A Study on STM Sosyal Bilimlerde Dönemsel Konu Modelleri: STM Üzerine Bir Çalişma


KURNAZ A., Unver H. A.

30th Signal Processing and Communications Applications Conference, SIU 2022, Safranbolu, Turkey, 15 - 18 May 2022 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu55565.2022.9864923
  • City: Safranbolu
  • Country: Turkey
  • Keywords: content analysis, social media, STM, text mining, topic models
  • Çanakkale Onsekiz Mart University Affiliated: Yes

Abstract

Topic models are rapidly becoming popular in social sciences. However, researchers should pay attention to some critical steps while using these models. The format and content of the textual data, language, existence of covariates, and preprocessing steps are the most crucial elements of a topic model analysis. This study inspects the effect of various datasets and preprocessing steps on Structural Topic Models (STM). Results shows that preprocessing, which depends on the research question, profoundly affects the model performance. Besides, the existence of multilingual data weakens the topic quality. Also, the algorithm performance is different among long and short texts. Last, the potential usage of covariates in the model enhances its functionality in social science.