Text embeddings and NLP for qualitative data analysis at scale

Contact person: Tor Ole Odden        
Keywords: natural language processing, large language models, qualitative data analysis, text embeddings, interdisciplinary    
Research groups: Center for Computing in Science Education (CCSE), Center for Interdisciplinary Education (INTED)    
Department of Physics

Qualitative analysis is a set of methods used across the social sciences—from education to sociology, political science, history, and psychology—in studies that deal with datasets that are difficult to quantize. Traditionally, these kinds of methods often require researchers to read through and classify large amounts of text data, like survey responses, interview transcripts, newspaper articles, or social media posts. However, recently-developed NLP analysis methods like text embeddings can potentially offer an alternative to this frequently opaque and time-intensive work, allowing researchers to use machine learning algorithms to replicably extract themes and classify or label large amounts of text data. Building on a base of both traditional qualitative research and NLP in science education, we are interested in developing and applying these methods: understanding how they work, where and how they can be used, their limitations, and how they can be combined with more traditional forms of qualitative analysis.

Methodological research topics:

  • Comparison of existing topic analysis methods (LDA, NMF, LSA) and text embeddings for unsupervised classification of text data
  • Automated literature reviews using NLP
  • Applications of text embeddings to analysis of interview transcripts or survey responses
  • Metrics for evaluation and benchmarking of NLP-aided qualitative research
  • Domain adaptation of large language models
  • Comparison between supervised and unsupervised algorithms for domain-specific text classification
  • Text and data cleaning using AI

Domains for NLP-assisted qualitative analysis:

  • Educational research and learning analytics, especially discipline-based education research (STEM education)
  • Environmental and sustainability politics and law
  • History
  • Political science
  • Psychology
  • Sociology and anthropology

External Partners:

  • Fridtjof Nansen Institute
  • Research groups across UiO departments of education, psychology, law, history, and political science    

Mentoring and internship will be offered by a relevant external partner.