Screening narratives of adolescents: A comparative analysis of BERTje, tf-idf, and multilingual DistilUSE feature extractors

Abstract

In psychological research, understanding identity formation involves screening and labeling textual data, such as narratives by teenagers about turning points in their lives. This process is similar to systematic screening for meta-analyses. Manual screening can be enhanced with AI-aided tools. This study explores active learning for analyzing Dutch narratives of teenagers’ turning points. A reliable feature extractor is needed to capture psychological, structural, and content dimensions in Dutch narrative identity data. The effectiveness of BERTje, tf-idf, and DistilUSE is compared for analyzing narrative identities among Dutch adolescents. BERTje performs best, demonstrating superior recall values and reduced average time to discover (ATD). BERTje shows promise for analyzing narrative data and extracting insights on identity formation. The performance gain may differ from screening abstracts. Showcasing BERTje’s effectiveness and the potential of AI-aided screening tools offer insights for comparative analysis and research on identity narratives. Integrating AI-aided tools in developmental psychology supports informed decision-making and enhances under- standing of narrative identities in interventions. These findings contribute to utilizing AI-aided tools in data analysis, understanding human experiences, and advancing psychological research and interventions.