quallmer: A New AI-Assisted Qualitative Analysis Tool in the R Language Ecosystem

quallmer is an open-source R package that provides researchers with a complete AI-assisted qualitative analysis toolbox. It supports large language model (LLM)-based automatic coding, cross-model reliability assessment, and auditable research workflows, helping researchers without programming or machine learning backgrounds conduct high-quality qualitative data analysis efficiently.

Qualitative research has long relied on manual coding and subjective interpretation, which is time-consuming and difficult to scale. With the improvement of LLM capabilities, the academic community is exploring the integration of AI into qualitative analysis workflows while maintaining research rigor and auditability. quallmer was born in this context, specifically designed for the R language, allowing researchers to use LLMs for qualitative analysis without in-depth programming or ML knowledge.

1. Intelligent Coding System (qlm_code): Automatically codes based on predefined codebooks, supports mainstream LLMs (GPT series/local open-source models), and returns structured objects to ensure reproducibility;
2. Custom Codebook (qlm_codebook): Creates coding schemes for specific studies and supports multiple data types;
3. Reliability Assessment: qlm_compare calculates consistency metrics, qlm_validate compares with manual gold standards, and qlm_replicate performs cross-model comparisons;
4. Audit Trail (qlm_trail): Generates complete records (model parameters, timestamps, archives, etc.) to ensure research credibility.

quallmer is built on the R language ecosystem and deeply integrated with tidyverse; it implements a unified LLM interface (cloud API/local deployment) through the ellmer package; it comes with the Shiny application quallmer.app, which provides an interactive interface (manual coding, AI annotation review, consistency calculation) to facilitate use by non-technical researchers.

Applicable to:

• Content analysis: Large-scale text topic coding;
• Sentiment analysis: Quickly obtaining sentiment tendency distributions;
• Multimodal research: Image/PDF document analysis;
• Mixed-methods research: Combining AI and manual coding to improve efficiency.

Usage Notes:

• AI coding is an aid, not a replacement for human judgment;
• Different LLMs have different coding styles, so cross-model validation is needed;
• For sensitive data, local deployment models are recommended;
• The quality of the codebook directly affects results, so careful design is required.

quallmer is an attempt to modernize qualitative research methodology. It encapsulates LLM capabilities into tools that comply with academic norms, provides auditable, verifiable, and reproducible workflows, and sets a new standard for AI-assisted qualitative research. It is applicable to fields such as social sciences and humanities, and future improvements in LLM capabilities will make such tools even more important.