همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

ChatGPT 4 and personality prediction

نویسندگان :

Zahra Eslami¹ Marcus Cheetham² Seyed Abolfazl Valizadeh³

1- Shahid Beheshti University 2- University Hospital Zurich 3- Shahid Beheshti University

کلمات کلیدی :

GPT4،Personality prediction،prompts styling

چکیده :

Large Language Models (LLMs) such as ChatGPT-4 have been successfully applied to infer the Big Five personality traits (openness, conscientiousness, extraversion, agreeableness, and neuroticism) from general text. However, their use for personality inference from LLM user interaction text (e.g., queries and retrieved information) remains underexplored. We hypothesize that conscientiousness acts as a moderator, increasing the predictive accuracy of LLM-inferred traits due to the linguistic alignment between characteristic conscientious text patterns and LLM training data. Participants (N=36) completed a validated self-report of the Big Five and used ChatGPT-4 for over 12 weeks. ChatGPT-4 was used to generate personality inferences from their text inputs. Correlational and moderation analyses revealed that the correlation between self-reported and LLM-inferred traits was significantly stronger for individuals with higher self-reported conscientiousness. This effect was most notable for the trait of conscientiousness itself, with modest, consistent moderation effects observed across the other Big Five traits. Predictive accuracy remained moderate overall, aligning with existing computational personality research. These findings suggest that the linguistic consistency of conscientious individuals facilitates more accurate LLM-based personality inference, emphasizing the need for improved model calibration and input assessment.