ObjectiveText analysis is a form of psychological assessment that involves converting qualitative information (text) into quantitative data. We tested whether automated text analysis using Generative Pre‐trained Transformers (GPTs) can match the “gold standard” of manual text analysis, even when assessing a highly nuanced construct like spirituality.MethodIn Study 1, N = 2199 US undergraduates wrote about their goals (N = 6597 texts) and completed self‐reports of spirituality and theoretically related constructs (religiousness and mental health). In Study 2, N = 357 community adults wrote short essays (N = 714 texts) and completed trait self‐reports, 5 weeks of daily diaries, and behavioral measures of spirituality. Trained research assistants and GPTs then coded the texts for spirituality.ResultsThe GPTs performed just as well as human raters. Human‐ and GPT‐generated scores were remarkably consistent and showed equivalent associations with other measures of spirituality and theoretically related constructs.ConclusionsGPTs can match the gold standard set by human raters, even in sophisticated forms of text analysis, but require a fraction of the time and labor.

中文翻译：

---

算法能告诉你有多属灵吗？使用 Generative Pretrained Transformers 进行复杂的文本分析


ObjectiveText 分析是一种心理评估形式，涉及将定性信息（文本）转换为定量数据。我们测试了使用生成式预训练转换器 （GPT） 的自动文本分析是否能与手动文本分析的“黄金标准”相匹配，即使在评估像灵性这样高度细微的结构时也是如此。方法在研究 1 中，N = 2199 名美国本科生写下了他们的目标（N = 6597 篇文本），并完成了灵性和理论相关结构（宗教和心理健康）的自我报告。在研究 2 中，N = 357 名社区成年人写了短文（N = 714 篇文本）并完成了特质自我报告、5 周的每日日记和灵性行为测量。然后，训练有素的研究助理和 GPT 对文本进行灵性编码。结果GPT 的表现与人类评分者一样好。人类和 GPT 生成的分数非常一致，并且与其他灵性测量和理论相关结构显示出等效的关联。结论GPT 可以匹配人类评分者设定的黄金标准，即使是在复杂的文本分析形式中，但需要的时间和劳动力只是其中的一小部分。