L4QAUCT HONOURS
2026
LLMs for Qualitative Analysis

Find the patterns.
Keep the context.

L4QA connects Grounded Theory, sentiment analysis and interactive visualisations so researchers can follow an insight back to the words that support it.

Explore the researchRead the poster
University of Cape Town · Computer Science · 2026
From source text to inspectable findings
01 / SOURCE MATERIAL

Interviews, transcripts & qualitative text

02 / DISCOVERGrounded TheoryCodes · categories · memos 02 / INTERPRETSentiment AnalysisDocuments · aspects · terms
Linked visualisationsExplore. Trace. Inspect. Refine.
One shared research questionHow can LLMs support qualitative analysis?
Three connected contributionsAnalysis, interpretation and evidence exploration
Researchers remain in the loopEvery result is open to inspection
01 / A connected approach

Three engines.
One evidence trail.

Each contribution tackles a different part of the qualitative research process. Together, they connect interpretation with the material behind it.

01

Grounded Theory

Specialist agents develop codes, categories and analytic memos. Auditing and targeted revision check the grounding of emerging interpretations.

Sibonelelo’s contribution
02

Sentiment Analysis

A comparison of unified and modular workflows for sentiment at document, aspect-category and aspect-term levels.

Sithokomele’s contribution
03

Visualisations

Linked views connect patterns to source excerpts, supporting selection, filtering, evidence inspection and human validation.

Theo’s contribution
02 / The research challenge

Automation is useful.
Interpretation still matters.

Qualitative analysis involves more than assigning a label to a passage. Researchers develop categories, consider context and continually revisit the evidence behind an interpretation.

L4QA investigates where LLM workflows can help, how specialised agents change the results, and how linked views keep researchers involved.

01

Support the analysis

Assess LLM workflows for Grounded Theory coding and multi-level sentiment analysis.

02

Compare the approaches

Examine the strengths and trade-offs of single-pass and multi-agent workflows.

03

Keep evidence accessible

Investigate linked visualisations for interpretation, traceability and human validation.

03 / Follow the evidence

A finding is a starting point.

Select a topic to see how a view can connect an interpretation with the words that support it.

Linked evidence explorerIllustrative example · invented excerpts
Select a topic

The selection changes the linked excerpt and interpretation together.

Illustrative interview ALines 12–13
“The first steps were clear, but I still wanted an example before coding on my own.”
Code: guided onboardingSentiment: mixed

Suggested interpretation

Clear instructions can coexist with a need for guided practice. The source contains both views.

1 of 2

This small interaction demonstrates the linking concept. It is not the research application and does not run an LLM.

04 / What the evaluation shows

Promising results.
Important trade-offs.

The project’s results make a case for targeted assistance and careful review. They do not show a uniform improvement on every measure.

Grounded Theory · category recall

54.5%

Up from 3.0% for the baseline. Code loss decreased, while semantic violations increased.

See the complete comparison

Sentiment Analysis · exact term F1

+4.2 pp

Term micro-F1 rises from 41.12% to 45.32%. Document sentiment macro-F1 declines slightly.

Explore the sentiment results

Visualisations · task completion

48 / 50

Tasks completed in a five-participant study. Mean error-spotting confidence was 2.8/5.

Read the study findings

Selected findings from the three final papers. Each contribution includes the study context, limitations and full paper.

05 / The people behind L4QA

Three perspectives.
A shared project.

Each individual contribution has its own research focus, method, evaluation and supporting documents.

Supervised by Gary StewartDepartment of Computer Science · University of Cape Town

Explore the project archive.

Project documents, final papers and website source code in one place.

Browse resources