LLM-assisted sentiment analysis for integrated computational and qualitative mixed methods education research: A case study of students' written reflection assignments

Researchers demonstrate how large language models can assist in analyzing student written reflections for mixed-methods education research, combining computational sentiment analysis with qualitative thematic analysis. The study of 151 study-abroad students reveals that prior international living experience significantly impacts sentiment toward language learning, suggesting LLM-assisted workflows enable efficient multi-variable demographic comparisons in qualitative research.

This research represents a meaningful intersection of educational methodology and AI capability deployment. Traditionally, qualitative analysis of student reflections has been labor-intensive, constraining researchers to single-variable comparisons across participant demographics. The study demonstrates that LLMs can process large volumes of written student data to surface sentiment patterns, which then guide deeper qualitative investigation. This workflow creates efficiency gains without sacrificing analytical rigor.

The broader context reflects growing adoption of LLMs as research infrastructure tools across academic disciplines. Rather than replacing human judgment, this application positions LLMs as accelerators for preliminary pattern recognition. The case study examined seven different student variables—including gender, cultural background, and prior international experience—finding that only previous abroad experience meaningfully influenced sentiment about communication development. This specificity suggests the methodology successfully filters signal from noise.

For education research and institutional learning analytics, this approach could democratize sophisticated mixed-methods analysis among resource-constrained institutions. Researchers without large coding teams could now extract comparable insights across multiple demographic dimensions. The implications extend to student success programs, where identifying which subpopulations respond differently to study-abroad experiences could inform targeted support interventions.

As LLM integration in research continues, attention must focus on validation protocols and transparency about model limitations. The study positions sentiment analysis as an entry point rather than final analysis, maintaining essential human interpretive work. Institutions exploring similar workflows should establish clear benchmarks for LLM performance and maintain detailed audit trails of analytical decisions.

• →LLM-assisted sentiment analysis enables efficient multi-variable demographic comparisons in qualitative education research previously limited by analysis time constraints.
• →Prior international living experience was the only significant variable affecting study-abroad students' sentiment toward language and communication development.
• →Mixed-methods workflows combining computational and qualitative analysis maintain research rigor while improving scalability across participant groups.
• →LLMs function most effectively as preliminary pattern-recognition tools guiding human qualitative investigation rather than replacing interpretive analysis.
• →This methodology could improve institutional capacity for learning analytics and personalized student support without proportional resource increases.