Language-based personality assessment from life narratives: a focus on model interpretability and efficiency

Traditional psychological assessments rely heavily on structured questionnaires, which may constrain nuanced self-expression. In this study, we investigate language-based personality assessment from open-ended life narratives (>2000 tokens) using large language models. We introduce a hybrid architecture combining sliding-window contextual embeddings with recurrent sequence modeling and attention mechanisms, focusing on both parameter efficiency and model interpretability. We benchmark against state-of-the-art architectures and demonstrate substantial gains in regression accuracy, computational throughput, and interpretability through attention weight visualization across Five-Factor Model (FFM) dimensions.

Frontiers in Artificial Intelligence, 9, 1760246