Elamur Research⌕
☰
home / papers / 2026 / Hierarchical User Intent Inference with Knowledge Graph Grounding

Hierarchical User Intent Inference with Knowledge Graph Grounding

Paper · 2026 · Tzu-Cheng Peng, Chien Chin Chen and Yung-Chun Chang

Summary

The paper addresses how to infer what hotel guests actually want from their reviews. That intent depends on explicit aspect judgments and on implicit context such as travel purpose, which plain LLMs, lacking structured knowledge, tend to miss. The authors propose HII-KG, a two-stage pipeline:

  • Stage 1: LLaMA-3.1-8B, tuned with LoRA adapters, reads a review and outputs six aspect scores as a text sequence. These scores rate the hotel's Rooms, Cleanliness, Service, Value, Sleep Quality and Location.
  • Stage 2: Entity linking connects the review to a hotel knowledge graph built from TripAdvisor data. Cypher queries then extract a relevant subgraph, which is turned into plain sentences. Llama3.1-8B-Cypher uses that context with few-shot prompting, without fine-tuning, to write a first-person intent statement.

On TripAdvisor data, HII-KG beat encoder-based baselines on aspect F1 and beat prompted and fine-tuned generators on intent metrics. The work suggests that light graph-derived context can make LLM review interpretation more specific and factually consistent. Evaluation, however, relies mostly on automatic metrics and English-only data.

Key points

  • Hotel-KG: built from 399,000 TripAdvisor reviews of 745 hotels.
    • Node types include Facility, RoomType, TravelPurpose, Aspect and Region.
    • Relations include HAS_FACILITY, PREFERS and MENTIONED_IN.
  • Aspect rating data: a curated set of 9,000 English TripAdvisor reviews with 1–5 scores for all six aspects.
  • Aspect results: HII-KG had the best F1 on every aspect against BERT, RoBERTa, BART and T5, with an average gain of +4.5%.
    • Value: 0.87; Sleep Quality: 0.85.
    • Service: 0.86 vs. 0.81 for the best baseline; Cleanliness: 0.91 vs. 0.88.
  • Intent generation: HII-KG scored 0.322 BLEU-4, 0.471 ROUGE-L and 0.684 BERTScore.
    • Against LLaMA3.1-8B without KG, this is +3.3 BLEU, +2.9 ROUGE-L and +1.5 BERTScore.
    • It also beat GPT-3.5, Gemini 1.5 Flash and fine-tuned BART.
  • Limitations stated by the authors: the data is English-only with manually curated types, human evaluation was limited, and the model size may hinder deployment.
  • Inconsistency in the paper: the backbone description varies. Stage 1 is described as LLaMA-3.1-8B with LoRA, yet Stage 2 calls Llama3.1-8B-Cypher the same backbone.

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

 Institutions

🎓National Taiwan UniversityTaipei · Taiwan29 works · 2,116 citations

Also on the byline: Taipei Medical University.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 1Research subjects and languages the work is about.Knowledge graphs & ontologies · 192 works

Cite

Cite as Tzu-Cheng Peng, Chien Chin Chen and Yung-Chun Chang. 2026. Hierarchical User Intent Inference with Knowledge Graph Grounding. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.285

@inproceedings{peng2026hierarchical,
  title = {{Hierarchical User Intent Inference with Knowledge Graph Grounding}},
  author = {Tzu-Cheng Peng, Chien Chin Chen and Yung-Chun Chang},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.285},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.285}
}

References

19 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Gaurav Arora, Shreya Jain, and Srujana Merugu. 2024. Intent detection in the age of llms. arXiv preprint arXiv:2410.01627. find
  3. Anna Bodonhelyi, Efe Bozkir, Shuo Yang, Enkelejda Kasneci, and Gjergji Kasneci. 2024. User intent recognition and satisfaction with large language models: A user study with chatgpt. arXiv preprint arXiv:2402.02136. — [arXiv:2402.02136](https://arxiv.org/abs/2402.02136) find
  4. Yung-Chun Chang, Chih-Hao Ku, and Chien-Hung Chen. 2020. Using deep learning and visual analytics to explore hotel reviews and responses. Tourism Management, 80:104129. find
  5. Yung-Chun Chang, Chih-Hao Ku, and Chun-Hung Chen. 2019. Social media analytics: Extracting and visualizing hilton hotel ratings and reviews from tripadvisor. International Journal of Information Management, 48:263–279. find
  6. Liyi Chen, Panrong Tong, Zhongming Jin, Ying Sun, Jieping Ye, and Hui Xiong. 2024. Plan-on-graph: Self-correcting adaptive planning of large language model on knowledge graphs. Advances in Neural Information Processing Systems, 37:37665–37691. find
  7. Jan Ole Krugmann and Jochen Hartmann. 2024. Sentiment analysis in the age of generative ai. Customer Needs and Solutions, 11(1):3. find
  8. Chih-Hao Ku, Yung-Chun Chang, and Yichuan Wang. 2024a. How to strategically respond to online hotel reviews: A strategy-aware deep learning approach. Information & Management, 61(5):103970. find
  9. Chih-Hao Ku, Yung-Chun Chang, and Yichuan Wang. 2024b. How to strategically respond to online hotel reviews: A strategy-aware deep learning approach. Information & Management, 61(2). find
  10. I-Fan Lin, Faegheh Hasibi, and Suzan Verberne. 2024. Generate then refine: data augmentation for zero-shot intent detection. arXiv preprint arXiv:2410.01953. find
  11. Chuangtao Ma, Yongrui Chen, Tianxing Wu, Arijit Khan, and Haofen Wang. 2025. Large language models meet knowledge graphs for question answering: Synthesis and opportunities. arXiv preprint arXiv:2505.20099. find
  12. Charaf Ouaddi, Lamya Benaddi, Lahbib Naimi, Mohamed Rahouti, Abdeslam Jakimi, Rachid Saadane, and 1 others. 2025. Assessing the effectiveness of large language models for intent detection in tourism chatbots: A comparative analysis and performance evaluation. Scientific African, 28:e02649. find
  13. Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2024. Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering, 36(7):3580–3599.
  14. Juan A Rodriguez, Nicholas Botzer, David Vazquez, Christopher Pal, Marco Pedersoli, and Issam Laradji. 2024. Intentgpt: Few-shot intent discovery with large language models. arXiv preprint arXiv:2411.10670. — [arXiv:2411.10670](https://arxiv.org/abs/2411.10670) find
  15. Shijie Wang, Wenqi Fan, Yue Feng, Lin Shanru, Xinyu Ma, Shuaiqiang Wang, and Dawei Yin. 2025. Knowledge graph retrieval-augmented generation for LLM-based recommendation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 27152–…
  16. Xinyang Xu, Zeyu Wang, Xinyi Li, Hang Dong, Qingqing Zheng, Yijia Liu, Wayne Xin Zhao, and Ji-Rong Wen. 2024a. Text2cypher: Training llms to generate executable cypher queries over knowledge graphs. arXiv preprint arXiv:2412.10064. find
  17. Zhentao Xu, Mark Jerome Cruz, Matthew Guevara, Tie Wang, Manasi Deshpande, Xiaofeng Wang, and Zheng Li. 2024b. Retrieval-augmented generation with knowledge graphs for customer service question answering. In Proceedings of the 47th international ACM SIGIR conference on research and development in in… find
  18. Dong Zhang and Baozhuang Niu. 2024. Leveraging online reviews for hotel demand forecasting: A deep learning approach. Information Processing & Management, 61(1):103527. find
  19. Wenqing Zheng, Noah Fatsi, Daniel Barcklow, Dmitri Kalaev, Steven Yao, Owen Reinert, C Bayan Bruss, and Daniele Rosa. 2025. Explain what you mean: Intent augmented knowledge graph recommender built with an llm. arXiv preprint arXiv:2505.10900. — [arXiv:2505.10900](https://arxiv.org/abs/2505.10900) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/peng-2026-hierarchical-user-intent-inference-knowledge/