In the presentation, we discuss a series of studies exploring innovative approaches to psychometric assessment that aim to move beyond traditional self-report questionnaires in the future. Our work integrates Large Language Model (LLM) embeddings and biosensor-based physiological monitoring to enhance personality assessment methodologies.
First, we examine how LLM-generated text embeddings can predict factor loadings in personality tests by analyzing the semantic structure of test items. Our studies on the Big Five personality test demonstrate that embeddings can effectively capture semantic similarities, providing a data-driven method for refining questionnaire design without relying on human response data. This approach offers a new perspective on evaluating and structuring psychological tests and move beyond self-rating scales towards the use of more natural language interactions.
Moreover, we present our work on biosensor-based assessment of response burden, to measure physiological signals during questionnaire completion. By recording participants' physiological responses while they engage with personality tests of different lengths, we investigate the cognitive and emotional load associated with survey completion. Our findings suggest that physiological measures can serve as objective indicators of response burden, complementing traditional self-report methodologies.
Through this presentation, we aim to share insights gained from these studies and discuss how the integration of LLMs and biosensors can improve psychometric research. These methods have the potential to enhance test design, reduce response biases, and pave the way for more robust and engaging psychological assessments.