An intelligent agent for sentence completion test: creation and application in depression assessment

During large-scale psychological screening, traditional self-report questionnaires face challenges like response deception or social desirability bias, while the Sentence Completion Test (SCT) as a projective technique shows potential but is limited by manual scoring and high costs. Leveraging advancements in Large Language Models (LLMs), this study integrates SCT's theoretical framework with LLM capabilities to develop a specialized set of SCT items for depression assessment in Chinese university students, using a self-built intelligent agent across three progressive empirical studies. Results show the agent demonstrates good reliability (Cronbach's <i>α</i> = 0.89-0.92) and validity, with high consistency to manual scoring (<i>r</i> = 0.96), significant criterion correlations with the Beck Depression Inventory (<i>r</i> = 0.89) and Self-Rating Depression Scale (<i>r</i> = 0.85), confirmed structural validity via exploratory factor analysis. Furthermore, the intelligent agent could identify most invalid responses (F1 = 0.94, Accuracy = 0.99, Precision = 0.99, Recall = 0.90). This research marks a key milestone in SCT's intelligent transformation, driving innovation in psychological assessment and offering new academic and practical pathways.