Year: 2026
Type: dissertation
Abstract: People face interviews many times in everyday life, including job interviews.Because interviews strongly affect people's future opportunities, HCI researchers developed interview training systems that enable users to practice mock interviews without a human partner.These systems analyze aspects of users' behavior during a mock interview, such as posture and responses, and present feedback to users.However, most systems limit feedback to quantitative metrics or issue lists.As a result, these systems cannot provide users with concrete examples of the performance that the system considers suitable.As a result, users often struggle to decide how to revise their responses and what direction to pursue.To address this issue, we develop an interview training system that uses a LLM-driven Alter-Ego AI agent that references a user's background information and behaves similarly to the user.Our system simulates a mock interview by having the user's Alter-Ego AI agent participate and respond to questions based on the user's background and context.To provide diverse response examples, the system generates multiple Alter-Ego AI agents that stay consistent with the user's background while varying personality traits.This approach produces multiple responses to the same question with different tones and vocabularies.Users can refer to these responses as concrete examples, incorporate effective elements into their own responses, and improve their interview responses.We conducted a user study in the context of job interview practice in order to evaluate the system.The results show that users can observe the Alter-Ego AI agent's responses from a third-person perspective and revise their own responses while considering the impressions created by specific phrases and topics.We also observed that unintended hallucinations from the large language model can prompt users to reflect on their past from perspectives ii Abstract that differ from their fixed assumptions.We discuss how this effect can support users' self-analysis.
Author: yamanaka, shun
Institution: The University of Tokyo
Language: English
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7208326117