[4K4-GS-6b] Language media processing

2:45 PM - 3:00 PM JST(5:45 AM - 6:00 AM UTC)

[4K4-GS-6b-06] Question-Guiding Strategy Based on Question Tree for Adaptive Interviews Balancing Self-Disclosure and Information Extraction

〇Fuminori Nagasawa¹, Ekai Hashimoto¹, Shun Shiramatsu¹ (1. Nagoya Institute of Technology)

Keywords:

Dialog System,Interview System,Large Language Models(LLMs)

To appropriately recommend functions and services through interview dialogue systems that adapt topic continuation/transition …, the system must guide question topics so that necessary information can be collected during topic adaptation. To achieve this, we propose a Question Tree—a tree-structured graph that organizes dynamically generated follow-up questions—and a question-guiding strategy that treats topic continuation/transition as graph exploration. Candidate questions generated by an LLM are added as nodes, and the next question is selected to approach predefined target information based on semantic similarity and the current topic-adaptation status. We conducted dialogue simulations with an LLM-based user and evaluated the target information collection rate under externally controlled topic continuation/transition scenarios.