Scholarlink

<p>Implicit motives provide a valuable construct for motivational research, that allows to explain diverse behaviours but comes with high manual effort for being assessed. In this work, we present an automatic approach to measure motive imagery in Picture Story Exercise (PSE) stories and running text. We compare two machine learning models, one based on the marker word approach and one grasping sentences more holistically with a sentence embedding, and find the latter to outperform the first on a dataset of unseen PSE stories, reaching a correlation with human coders of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(&gt;.80\)</EquationSource> </InlineEquation> on persons’ motives. We further demonstrate the embedding model’s capabilities by showing correlations of the automatically encoded motive scores with related behavioural outcome criteria on large amounts of unseen longitudinal data from the National Child Development Study. Among others, motives encoded by the model at participant’s age eleven are associated with sports activity, academic achievement and voluntary work years and decades later. Our research raises hopes of using machine learning to encode implicit motives in written text, which would facilitate further research by lowering manual effort of implicit motive assessment by orders of magnitude.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Encoding implicit motives in text: using machine learning for automated assessment of PSE stories and running text

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Dorian Drost,
  • Marco Stojanovic,
  • Stefan Fries

摘要

Implicit motives provide a valuable construct for motivational research, that allows to explain diverse behaviours but comes with high manual effort for being assessed. In this work, we present an automatic approach to measure motive imagery in Picture Story Exercise (PSE) stories and running text. We compare two machine learning models, one based on the marker word approach and one grasping sentences more holistically with a sentence embedding, and find the latter to outperform the first on a dataset of unseen PSE stories, reaching a correlation with human coders of \(>.80\) on persons’ motives. We further demonstrate the embedding model’s capabilities by showing correlations of the automatically encoded motive scores with related behavioural outcome criteria on large amounts of unseen longitudinal data from the National Child Development Study. Among others, motives encoded by the model at participant’s age eleven are associated with sports activity, academic achievement and voluntary work years and decades later. Our research raises hopes of using machine learning to encode implicit motives in written text, which would facilitate further research by lowering manual effort of implicit motive assessment by orders of magnitude.