Elamur Research⌕
☰
home / papers / 2026 / Tree-Based Interview Topic Guidance for Collecting Target Information under Adaptive Topic continuation/switching

Tree-Based Interview Topic Guidance for Collecting Target Information under Adaptive Topic continuation/switching

Summary

The paper tackles a tension in interview-style dialogue systems for service recommendation. Adapting topics to the user's mood encourages self-disclosure, but conversations then drift toward whatever the user prefers to discuss, and the system misses information it needs. The authors propose a Question Tree, in which questions are nodes connected by parent–child "follow-up" links. After each user answer, an LLM produces two kinds of candidate questions. Deepening questions probe the current topic further, and approach questions steer toward the target information. The candidates are filtered for naturalness and added to the tree. An external module decides whether to continue or switch topics, and the next question is then picked from the current node's children or siblings. Ranking combines sentence-embedding similarity to the dialogue context and to the target. In Japanese-language simulations with an LLM dummy user and randomly controlled continuation/switching, the method raised target-information collection and made it more consistent across sessions. A small human evaluation found no significant loss of naturalness. This makes tree-based guidance a plug-in complement to existing topic-adaptation modules.

Key points

  • Mean target-information collection rate after 10 turns rose from 10.8% (baseline) to 26.6% with the combined method, and variability between sessions decreased.
  • Candidate ranking uses a weighted blend, S_blended = r_c·S_c + r_t·S_t. Here S_c is similarity to the current context and S_t is similarity to the target. With several targets, the highest-scoring candidate across all of them is selected.
  • Deepening questions are diversified using 5W1H and qualitative-interview probe types. At least one approach question is generated for each target item.
  • The experiments used 24 target-information items, translated from Japanese, such as hobbies, days off, flaws and painful experiences. These items are graded by self-disclosure level from 1 to 4.
  • Mean processing time per turn was about 4.8 s (range roughly 3.5–6.3 s), and question-candidate generation was the largest component at about 2.4 s.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.sigdial-1.35

@inproceedings{nagasawa2026tree,
  title = {{Tree-Based Interview Topic Guidance for Collecting Target Information under Adaptive Topic continuation/switching}},
  author = {Nagasawa, Fuminori and Hashimoto, Ekai and Shiramatsu, Shun},
  booktitle = {Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue},
  year = {2026},
  doi = {10.18653/v1/2026.sigdial-1.35},
  url = {https://doi.org/10.18653/v1/2026.sigdial-1.35}
}

References

22 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. In this study, we developed a Japanese-language dialogue system and employed a self-disclosure scale tailored specifically for Japanese. When con- References ducting dialogue experiments with actual human Irwin Altman. 1973. Social penetration: The devel-participants, it is necessary to be mindful o… find
  2. Chloé Clavel, Matthieu Labeau, and Justine Cassell. 2022. Socio-conversational systems: Three challenges at the crossroads of fields. Frontiers in Robotics and AI, 9. find
  3. Harm de Vries, Florian Strub, Sarath Chandar, Olivier Pietquin, Hugo Larochelle, and Aaron Courville. 2017. Guesswhat?! visual object discovery through multi-modal dialogue. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). find
  4. Ben Emans. 2016. Interviewing: Theory, techniques and training. Routledge, London. find
  5. Andrea Fontana and James H Frey. 2005. The interview. The Sage handbook of qualitative research, 3(1):695– 727. find
  6. Chuanruo Fu and Yuncheng Du. 2025. First ask then answer: A framework design for ai dialogue based on supplementary questioning with large language models. arXiv preprint arXiv:2508.08308. — [arXiv:2508.08308](https://arxiv.org/abs/2508.08308) find
  7. Ekai Hashimoto, Mikio Nakano, Takayoshi Sakurai, Shun Shiramatsu, Toshitake Komazaki, and Shiho Tsuchiya. 2025. A career interview dialogue system using large language model-based dynamic slot generation. In Proceedings of the 31st International Conference on Computational Linguistics, pages 1562– 1… find
  8. Conference on Empirical Methods in Natural Language Processing, pages 12448–12460, Singapore. Association for Computational Linguistics. find
  9. Fuminori Nagasawa and Shogo Okada. 2025. Can Adaptive Interviewer Robot Based on Social Signals Make a Better Impression on Interviewees and Encourage Self-Disclosure?, page 35–43. Association for Computing Machinery, New York, NY, USA. find
  10. Sora Niwa and Shun’ichi MARUNO. 2010. Development of a scale to assess the depth of self-disclosure. Japanese Journal of Personality, 18(3). find
  11. OpenAI. 2025. gpt-oss-120b & gpt-oss-20b model card. Preprint, arXiv:2508.10925. find our card
  12. Hashai Papneja and Nikhil Yadav. 2024. Self-disclosure to conversational ai: A literature review, emergent framework, and directions for future research. Personal and ubiquitous computing, pages 1–33. find
  13. Jeffrey M Stanton. 2021. Evaluating equivalence and confirming the null in the organizational sciences. Organizational Research Methods, 24(3):491–512. find
  14. Shohei Tanaka, Koichiro Yoshino, Katsuhito Sudoh, and Satoshi Nakamura. 2021. Arta: Collection and classification of ambiguous requests and thoughtful actions. Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue. find
  15. Ryuichiro Higashinaka, Tetsuro Takahashi, Sota Horiuchi, Michimasa Inaba, Shiki Sato, Kotaro Funakoshi, Masato Komuro, Hiroyuki Nishikawa, Mayumi Usami, Takashi Minato, Kurima Sakai, and Tomo Funayama. 2022. Dialog system live competition 5. JSAI SIG-SLUD, 96:19. find
  16. Yuki Zenimoto, Mariko Yoshida, Ryo Hori, Mayu Urada, Aiko Inoue, Takahiro Hayahsi, and Ryuichiro Higashinaka. 2025. Development of a question-guided survey dialogue system. Proceedings of the 31st Annual Conference of the Association for Natural Language Processing. find
  17. Allison Hurst. 2023. Interviewing. Introduction to Qualitative Research Methods. find
  18. Tiancheng Zhao and Maxine Eskenazi. 2016. Towards end-to-end learning for dialog state tracking and management using deep reinforcement learning. In Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 1–10, Los Angeles. Association for Computational… find
  19. Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. 2022. Clam: Selective clarification for ambiguous questions with generative language models. arXiv preprint arXiv:2212.07769. find
  20. Shikib Mehri and Maxine Eskenazi. 2020. Unsupervised evaluation of interactive dialog with DialoGPT. In Proceedings of the 21st Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 225–235, 1st virtual meeting. Association for Computational Linguistics. find our card
  21. Seiya Mitsuno, Midori Ban, Hiroshi Ishiguro, and Yuichiro Yoshikawa. 2024. Deepening conversations over time: A chatbot with a topic depth estimation model for gradually engaging in deeper chats. In 2024 33rd IEEE International Conference on Robot and Human Interactive Communication (ROMAN), pages 1… find
  22. John Morris, Volodymyr Kuleshov, Vitaly Shmatikov, and Alexander Rush. 2023. Text embeddings reveal (almost) as much as text. In Proceedings of the 2023 find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nagasawa-2026-tree-based-interview-topic-guidance/