Real-Time Control of NPC Behavior Using Learning Models Based on the Big Five Personality Traits

Abstract

In recent years, while AI technology in the gaming industry has advanced to enhance gameplay and player experience, challenges such as dynamic agent configuration and the cost of retraining remain. This study proposes a method that adjusts the output of pre-trained models based on probability distributions, enabling the generation of agents with diverse personality settings without requiring retraining. This approach aims to reduce development costs and training time, as well as support the design of personality expressions through dynamic personality configuration changes during gameplay.