Akademik Veri Yönetim Sistemi
2nd INTERNATIONAL WRITETEC CONGRESS OF SOCIAL SCIENCES AND HEALTH SCIENCES IN THE AGE OF ARTIFICIAL INTELLIGENCE, İstanbul, Türkiye, 21 - 23 Şubat 2025, ss.35, (Özet Bildiri)
Access to data sources in social sciences, especially business science, has become problematic in recent years. In all studies where the data source is human, the objectivity and impartiality of the data may be under suspicion. The use of videos and transcripts in these videos is a new research area that has the potential to solve this situation. Science always contains hidden information and deep meanings that cannot be obtained by quantitative methods and should be investigated by qualitative methods (Wach, Stephan, & Gorgievski, 2016, p. 1113; Yılmaz, Karaman, & Cinar, 2013, p. 155). Qualitative research has been criticized for the potential subjective bias of participants and practitioners at the point of obtaining data while trying to reach social reality by its nature. Conducting qualitative research with video data as in this study is a new phenomenon (Heath, Hindmarsh, & Luff, 2010). Large language models (LLM), one of the artificial intelligence tools, can be a solution to the problems of video-based qualitative research (Abram, Mancini, & Parker, 2020). In this context, using the video-based research method (Ormiston & Thompson, 2021, p. 976), which is a new research phenomenon for social sciences, the “entrepreneurial success” themed videos obtained from the channel named “StoryBox” broadcasting on YouTube between 2020-2024 were analyzed. This study aims to use big language models (LLM) and natural language processing (NLP) to gain in-depth insights into qualitative research themes and categories, the whys and hows of social phenomena. In this study, the transcripts of the videos were extracted with “Youtube API”. The analysis process of the study consists of text preprocessing and thematic feature extraction, revealing the semantic network structure by optimizing modules, conceptual categorization with Zero-shot classification, and N-gram-based subcategory discovery. In the findings part of the study, 18 categories were obtained in 3 thematic areas related to interventional success. This study is considered to be valuable in terms of its potential for dissemination and diversification in many fields of social and health sciences. Especially in terms of implicit or explicit knowledge transfer of the institutional and scientific memory of professionals in health and social sciences and knowledge transfer for future research, this research model is critical in transferring not only technical knowledge but also intuitive decision-making mechanisms and problem-solving strategies.