A Large Language Model-based Self-analysis Support Dialogue System for Job-hunting Students

To prevent mismatches in job hunting, self-analysis is widely encouraged. However, conventional methodssuch as worksheets require high metacognitive ability and tend to focus on the“ present self ”, making it difficult forstudents to examine their decisions in relation to future goals and life design. This study introduces a time perspectivein which students reflect on the present from the standpoint of their future selves, and proposes a dialogue systemthat supports self-analysis by prompting users to notice inconsistencies between their current values and decisionsmade after concretizing a future vision. The system asks job-hunting students to rank evaluation criteria for companyselection as a values list, elicits their future dreams, and then generates staged questions about the consistency betweentheir dreams and current values, as well as the possibility that their values may change after achieving those dreams.It was implemented as a web application with an adaptive probing mechanism based on utterance classification andemotion analysis using voice emotion labels and response intervals. A comparative experiment with 15 engineeringundergraduate job-hunting students showed medium-to-large effect sizes for clarifying work-career and private-lifegoals and for gaining new perspectives, such as work-career goals (Cliff ’s δ = 0.58) and gaining a new perspective(Cliff ’s δ = 0.50). The amount of change in ranked values also tended to increase in the latter half of the dialogue.Given the small sample size, future work should validate the system with a larger, more diverse population and furtherrefine the intervention strategy.