PKSHA Technology and Tohoku University have jointly researched incorporating cognitive styles into LLM-based dialogue simulations for persuasion. The study analyzed how cognitive traits like need for cognition, need for cognitive closure, and maximization tendency affect dialogue behavior. Unlike previous simulations focusing primarily on personality traits, this research evaluated how information perception and decision-making styles impact responses. Using data from over 2,000 real individuals, the traits were integrated into LLM user simulators. Results showed significant changes in dialogue behavior, such as increased utterances examining conditions and a higher tendency to accept persuasion when conditions are met. A significant correlation between cognitive style strength and utterance tendencies was also confirmed. The research aims to improve PKSHA AI Agents by verifying dialogue content tailored to users' cognitive styles. The findings will be presented at the Japanese Society for Artificial Intelligence Annual Conference (JSAI2026) poster session on June 10, 2026.

FACT BOX

  • Source: PR TIMES
  • Category: business