Large Language Models for Psychological Assessment: A Comprehensive Overview (journals.sagepub.com)

This paper offers a practitioner-focused overview showing how large language models (LLMs) can be used as scalable, multimethod tools for psychological assessment. It argues that language — unlike traditional self-report questionnaires — is behavioral, ecologically valid, and rich enough to capture multiple constructs from a single sample, potentially reducing social desirability, recall, and cultural biases. The authors map practical benefits (faster, broader, deployable in low-resource or emergency settings), outline experiment-design best practices, and supply reproducible resources: a GitHub tutorial, code examples, and a glossary to help researchers and clinicians get started. Technically, the review traces the evolution from bag-of-words tools (e.g., LIWC) through static embeddings (Word2vec, GloVe) and RNN/LSTM models to transformer-based architectures whose self-attention mechanism enables context-sensitive representations across long texts. It summarizes encoder-only, decoder-only, and encoder–decoder families (e.g., BERT, GPT-style, T5/BART), notes the impact of model scaling (millions → hundreds of billions of parameters), and highlights implications: LLMs can infer many constructs from language but require rigorous construct validation, careful experimental design, ethical oversight, and implementation safeguards. The paper’s code and methodological guidance aim to accelerate responsible adoption of LLM-based assessment in research and clinical practice.