The question
How do you find themes and sentiment in hundreds of interview responses without reading and coding every single one by hand?
What the data showed
The app grouped responses with the same meaning and showed the emotional tone. Psychologists and researchers could explore the groups interactively and compare several language models in the same tool.
What it could be used for
An overview of large amounts of qualitative data as a starting point for interpretation. The human still makes the judgment, but starts from a whole picture instead of a pile of text.
Tools
R Shiny and reticulate running sentence-transformers. UMAP and clustering (K-means / HDBSCAN). Interactive visualization with plotly and ggplot2, and benchmarking on multilingual data.
The same tool can group and make sense of any free text — survey answers, customer enquiries, reviews, or open responses from employees.