Researchers propose a fine-tuned multi-agent framework to detect OCEAN personality traits from life narratives, addressing the challenge of latent and context-dependent traits in text. The system uses sub-agents conditioned via masked language modeling and psychometric supervision to adopt high, low, or neutral perspectives for each trait.
- Sub-agents are conditioned to adopt specific perspectives (high, low, or neutral) for each OCEAN trait using masked language modeling and psychometric supervision.
- A judge LLM aggregates and compares the outputs from these sub-agents to generate final trait predictions, mitigating individual model biases.
- The framework is evaluated on a life narrative dataset through quantitative and qualitative experiments, including baselines and ablations.
The approach provides a scalable and interpretable method for text-based personality inference by leveraging multi-agent reasoning grounded in psychometric supervision.