Skip to contents

Identify low-quality open-ended responses using the OpenAI API

Source: R/06_02_data_quality_lowqual_gpt.R
lowqual_gpt.Rd

This function flags low-quality open-ended survey responses using the OpenAI API. Responses are evaluated for signs of gibberish, nonsense, random text, irrelevant words, or other indicators of poor data quality.

Usage

lowqual_gpt(data, x, batch_size = 100, model = "gpt-4o")

Arguments

data

A data frame containing the survey data.

x

The open-ended variable to evaluate for response quality.

batch_size

Integer; number of responses per API call. Defaults to 100.

model

Character string; the OpenAI model to use. Defaults to gpt-4o.

Value

A data frame with an additional variable named x_lowqual containing the classification results.

Details

Requires an OpenAI API key, which can be generated at https://platform.openai.com/, to be set in your R session using Sys.setenv(OPENAI_API_KEY="...").

The model classifies each response as low-quality (1) or valid (0).

A response is flagged as low-quality if it is:

  • Gibberish or random characters

  • Off-topic or meaningless

  • Contains only emojis or irrelevant text

A response is considered valid if it is interpretable, relevant, and meaningful.

The function appends a new column to the dataset with the results.

Examples

# Identify low-quality respondents in question Q5
data <- lowqual_gpt(data = survey_data, x = Q5)
#> Error: These packages are required but not installed: dplyr, stringr