使用大型语言模型和人工角色的人格测试开发的初始阶段的框架

IF 3.1 2区心理学 Q2 PSYCHOLOGY, SOCIAL

Journal of Research in Personality Pub Date : 2025-07-31 DOI:10.1016/j.jrp.2025.104647

Patrick M. Markey, Hanna Campbell, Samantha Goldman

{"title":"使用大型语言模型和人工角色的人格测试开发的初始阶段的框架","authors":"Patrick M. Markey, Hanna Campbell, Samantha Goldman","doi":"10.1016/j.jrp.2025.104647","DOIUrl":null,"url":null,"abstract":"<div><div>This study explored using Large Language Models (LLMs) in early personality test construction, presenting a method to efficiently assess item relevance to psychological constructs. Study 1 generated self-esteem and Five-Factor Model (FFM) scales by analyzing AI-agent responses, resulting in scales with high internal consistency and face validity. Study 2 tested these scales with 449 human participants, finding that the AI-created self-esteem scale showed satisfactory internal consistency and strong correlations with the Rosenberg Self-Esteem Scale. The AI-created FFM scales demonstrated satisfactory internal consistency, convergent and divergent validity with the NEO Personality Inventory-Revised, and similar correlational patterns, though with some discrepancies in Agreeableness. These findings suggest LLMs can streamline item selection in personality test development.</div></div>","PeriodicalId":48406,"journal":{"name":"Journal of Research in Personality","volume":"118 ","pages":"Article 104647"},"PeriodicalIF":3.1000,"publicationDate":"2025-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A framework for the initial phases of personality test development using large language models and artificial personas\",\"authors\":\"Patrick M. Markey, Hanna Campbell, Samantha Goldman\",\"doi\":\"10.1016/j.jrp.2025.104647\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>This study explored using Large Language Models (LLMs) in early personality test construction, presenting a method to efficiently assess item relevance to psychological constructs. Study 1 generated self-esteem and Five-Factor Model (FFM) scales by analyzing AI-agent responses, resulting in scales with high internal consistency and face validity. Study 2 tested these scales with 449 human participants, finding that the AI-created self-esteem scale showed satisfactory internal consistency and strong correlations with the Rosenberg Self-Esteem Scale. The AI-created FFM scales demonstrated satisfactory internal consistency, convergent and divergent validity with the NEO Personality Inventory-Revised, and similar correlational patterns, though with some discrepancies in Agreeableness. These findings suggest LLMs can streamline item selection in personality test development.</div></div>\",\"PeriodicalId\":48406,\"journal\":{\"name\":\"Journal of Research in Personality\",\"volume\":\"118 \",\"pages\":\"Article 104647\"},\"PeriodicalIF\":3.1000,\"publicationDate\":\"2025-07-31\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Research in Personality\",\"FirstCategoryId\":\"102\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0092656625000790\",\"RegionNum\":2,\"RegionCategory\":\"心理学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"PSYCHOLOGY, SOCIAL\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Research in Personality","FirstCategoryId":"102","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0092656625000790","RegionNum":2,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"PSYCHOLOGY, SOCIAL","Score":null,"Total":0}

引用次数: 0

摘要

本研究探讨了在早期人格测验建构中使用大语言模型，提出了一种有效评估项目与心理建构相关性的方法。研究1通过分析人工智能主体的反应生成自尊和五因素模型（FFM）量表，得到具有较高内部一致性和面孔效度的量表。研究2对449名人类参与者进行了测试，发现人工智能自尊量表与罗森博格自尊量表具有令人满意的内部一致性和强相关性。人工智能创建的FFM量表显示出令人满意的内部一致性，与NEO人格量表修订版的收敛效度和发散效度，以及类似的相关模式，尽管在宜人性方面存在一些差异。这些发现表明llm可以简化人格测试开发中的项目选择。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

A framework for the initial phases of personality test development using large language models and artificial personas

This study explored using Large Language Models (LLMs) in early personality test construction, presenting a method to efficiently assess item relevance to psychological constructs. Study 1 generated self-esteem and Five-Factor Model (FFM) scales by analyzing AI-agent responses, resulting in scales with high internal consistency and face validity. Study 2 tested these scales with 449 human participants, finding that the AI-created self-esteem scale showed satisfactory internal consistency and strong correlations with the Rosenberg Self-Esteem Scale. The AI-created FFM scales demonstrated satisfactory internal consistency, convergent and divergent validity with the NEO Personality Inventory-Revised, and similar correlational patterns, though with some discrepancies in Agreeableness. These findings suggest LLMs can streamline item selection in personality test development.

求助全文

通过发布文献求助，成功后即可免费获取论文全文。去求助

来源期刊

Journal of Research in Personality PSYCHOLOGY, SOCIAL-

CiteScore

5.40

自引率

6.10%

发文量

102

审稿时长

67 days

期刊介绍： Emphasizing experimental and descriptive research, the Journal of Research in Personality presents articles that examine important issues in the field of personality and in related fields basic to the understanding of personality. The subject matter includes treatments of genetic, physiological, motivational, learning, perceptual, cognitive, and social processes of both normal and abnormal kinds in human and animal subjects. Features: • Papers that present integrated sets of studies that address significant theoretical issues relating to personality. • Theoretical papers and critical reviews of current experimental and methodological interest. • Single, well-designed studies of an innovative nature. • Brief reports, including replication or null result studies of previously reported findings, or a well-designed studies addressing questions of limited scope.