text-classification

What problem does it solve?

Social science researchers often struggle to design and document rigorous, reproducible LLM-based text classification workflows. This Skill provides a structured blueprint for building such workflows, including codebook design, learning-regime selection, piloting, and transparent reporting.

Core Features & Use Cases

• Codebook design guidance with five components per code: Label, Definition, Clarification, Negative clarification, and Examples.
• Guidance on learning regimes (zero-shot, few-shot, fine-tuning, instruction-tuning) and model selection (open-weight vs proprietary) to optimize classification tasks.
• Pilot testing and validation workflows against human ground truth, including inter-coder reliability considerations.
• Hybrid human-LLM workflows for uncertain classifications and robust reporting practices.
• Guidance on structured prompts and reproducibility documentation for publishable results.

Quick Start

Define your coding scheme using the five-code structure, choose a learning regime, run a 50–100 response pilot against human ground truth, and document the full pipeline.