[1Win4] Poster session 1

[1Win4-28] Agent Generation Method with Person Profile based on Behavioural History

〇Rikuto YOSHIZAWA Yoshizawa¹, Daichi Ojiro¹, Shigenori Matsumoto¹ (1.Hitachi)

Keywords:Agent, Simulation, Behavioral history, Purchase history, Retail

In recent years, there has been an increase in research into various methods of using large language models (LLMs) to generate agents that mimic personality. Among these, agents based on personal profiles generated from an individual's behavioral history are expected to be used in the marketing field. In the marketing field, providing timely services based on consumer experience is crucial; however, static profile generation methods based on traditional approaches struggle to meet these needs. To address this, we propose a method for continuously updating user profiles along a behavioral timeline and a memory recall technique that mimics decision-making based on past experience to generate agents. Using this method, we created customer agents based on purchase histories from retail stores and simulated purchase behaviour. The results showed that the proposed method achieved 67.7% accuracy in selecting products actually purchased, which is 4.3% higher than conventional methods.