Automated coding
Automated coding means letting software apply a research team's codes to qualitative material, through keyword rules, a trained classifier or a language model working from a codebook. Researchers still write the codes and check the output, while the software does the repetitive labeling across thousands of responses.
How it works
The simplest form searches for a word or phrase and applies a code wherever it appears. Machine-learning approaches go further: a model is trained on a sample coded by researchers, or given the codebook, and then applies the codes to the rest of the data. Recent natural language processing lets models read context and meaning rather than keywords alone, which improves accuracy on messy, conversational text.
Automation handles volume; researchers keep the judgment. The team defines the codes from the research question, checks a sample of the machine's coding against its own, and refines the codebook where the two disagree.
A worked example
A health insurer collects 18,000 open-ended comments from its member survey. Two researchers code 600 comments by hand against a 14-code codebook, and a model trained on those codes labels the rest. On a check sample of 300, the model agrees with the researchers on 88 percent of codes.
Related
Automate this in Pulse
Automated coding is easiest to trust when a person can spot-check it, and every quote Pulse's analysis files under a theme can be opened in its transcript to confirm the label fits.