Elamur Research⌕
☰
home / papers / 2026 / Hierarchical Reason-of-Contact Detection in Retail Banking Customer Interactions via LLM-Driven Taxonomy Induction

Hierarchical Reason-of-Contact Detection in Retail Banking Customer Interactions via LLM-Driven Taxonomy Induction

Paper · 2026 · Yoages Kumar Mantri, Yagneswaran Sriraja, John Hartley et al.

Summary

NatWest AI Research addresses Reason-of-Contact (RoC) detection, which means identifying why a customer contacted a retail bank. Existing intent schemes tend to be fragmented, maintained by hand, and quickly outdated as business processes change. The authors propose a two-stage framework:

  • Offline stage: about 50,000 historical chat and voice conversations are cleaned, PII-masked and split into atomic utterances. These are embedded with Sentence-BERT and clustered with UMAP+HDBSCAN. An LLM then extracts intents from each cluster and arranges them into a three-level (L0-L1-L2) taxonomy, which subject matter experts refine iteratively.
  • Online stage: live conversations are matched against taxonomy node embeddings using nearest-neighbour search (FAISS). The system produces a First RoC from early utterances for routing and an Overall RoC from a sliding window for analytics. It also enforces hierarchy consistency and flags out-of-domain cases.

The deployed system processes roughly 80,000 interactions daily with 150–300 ms latency. On a reference dataset it improves F1 by up to 10% over baselines. Because LLMs are kept out of real-time inference, the design stays deterministic, auditable and cheap, which suits regulated banking.

Key points

  • The taxonomy spans 14 L0 categories and more than 200 L2 intents, with descriptions at every node. Rare topics are grouped into a General Inquiries bucket.
  • Embeddings come from all-MiniLM-L6-v2 (384-dimensional), chosen from ten candidate models. Only the first ten customer utterances of each conversation are used for taxonomy induction.
  • Out-of-domain detection labels a conversation UNKNOWN when its best similarity score falls below 0.3. If L2 and L1 predictions conflict, the L2 prediction overrides the parent labels.
  • When new services were introduced, an ad-hoc taxonomy update was completed within two weeks without disrupting existing processes.
  • Deployment lessons: business requests for very fine-grained or department-duplicated intents hurt accuracy. The team resolved this by standardising duplicated intents as global intents and routing via downstream metadata.

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, 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 worksBusiness, Management and Accounting · 32 worksMarketing · 4 works

Cite

Cite as Yoages Kumar Mantri, Yagneswaran Sriraja, John Hartley, Greig A Cowan, Raad Khraishi and Abhinav Gupta. 2026. Hierarchical Reason-of-Contact Detection in Retail Banking Customer Interactions via LLM-Driven Taxonomy Induction. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.108

@inproceedings{mantri2026hierarchical,
  title = {{Hierarchical Reason-of-Contact Detection in Retail Banking Customer Interactions via LLM-Driven Taxonomy Induction}},
  author = {Yoages Kumar Mantri, Yagneswaran Sriraja, John Hartley, Greig A Cowan, Raad Khraishi and Abhinav Gupta},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.108},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.108}
}

References

21 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mihael Ankerst, Markus M. Breunig, Hans-Peter Kriegel, and Jörg Sander. 1999. Optics: ordering points to identify the clustering structure. SIGMOD Rec., 28(2):49–60. find
  2. Aymeric Beauchamp, Thi-Bich-Hanh Dao, Samir Loudni, and Christel Vrain. 2023. Incremental Constrained Clustering by Minimal Weighted Modification. In 29th International Conference on Principles and Practice of Constraint Programming (CP 2023), volume 280 of Leibniz International Proceedings in Infor… find
  3. Iñigo Casanueva, Tadas Temčinas, Daniela Gerz, Matthew Henderson, and Ivan Vulić. 2020. Efficient intent detection with dual sentence encoders. Preprint, arXiv:2003.04807.
  4. Ajay Chatterjee and Shubhashis Sengupta. 2020. Intent mining from past conversations for conversational agent. In Proceedings of the 28th International Conference on Computational Linguistics, pages 4140– 4152, Barcelona, Spain (Online). International Committee on Computational Linguistics. find
  5. Arshit Gupta, John Hewitt, and Katrin Kirchhoff. 2019. Simple, fast, accurate intent classification and slot labeling for goal-oriented dialogue systems. In Proceedings of the 20th Annual SIGdial Meeting on Discourse and Dialogue, pages 46–55, Stockholm, Sweden. Association for Computational Linguis… find
  6. Kurt Hornik, Ingo Feinerer, Martin Kober, and Christian Buchta. 2012. Spherical k-means clustering. Journal of Statistical Software, 50(10):1–22. find
  7. Kalervo Järvelin and Jaana Kekäläinen. 2002. Cumulated gain-based evaluation of ir techniques. ACM Trans. Inf. Syst., 20(4):422–446. find
  8. Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2017. Billion-scale similarity search with gpus. Preprint, arXiv:1702.08734.
  9. Jihyun Lee and Gary Lee. 2025. The limits of post-hoc preference adaptation: A case study on DSTC12 clustering. In Proceedings of the Twelfth Dialog System Technology Challenge, pages 36–43, Avignon, France. Association for Computational Linguistics. find
  10. Junhua Liu, Yong Keat Tan, Bin Fu, and Kwan Hui Lim. 2024. Lara: Linguistic-adaptive retrieval-augmentation for multi-turn intent classification. Preprint, arXiv:2403.16504. find
  11. Pengfei Liu, Youzhang Ning, King Keung Wu, Kun Li, and Helen Meng. 2021. Open intent discovery through unsupervised semantic clustering and dependency parsing. Preprint, arXiv:2104.12114. — [arXiv:2104.12114](https://arxiv.org/abs/2104.12114) find
  12. Leland McInnes, John Healy, and Steve Astels. 2017. hdbscan: Hierarchical density based clustering. Journal of Open Source Software, 2(11):205.
  13. Leland McInnes, John Healy, and James Melville. 2020. Umap: Uniform manifold approximation and projection for dimension reduction. Preprint, arXiv:1802.03426.
  14. A Padmasundari and Srinivas Bangalore. 2018. Intent discovery through unsupervised semantic text clustering. In Interspeech 2018, pages 606–610. find
  15. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. Preprint, arXiv:1908.10084.
  16. SBERT Team. 2023. Sentencetransformers utility functions documentation. https://sbert.net/docs/ package_reference/util.html. Accessed July 8, 2025. ↗
  17. Marina Sokolova and Guy Lapalme. 2009. A systematic analysis of performance measures for classification tasks. Information Processing Management, 45(4):427–437. find
  18. Mengting Wan, Tara Safavi, Sujay Kumar Jauhar, Yujin Kim, Scott Counts, Jennifer Neville, Siddharth Suri, Chirag Shah, Ryen W White, Longqi Yang, Reid Andersen, Georg Buscher, Dhruv Joshi, and Nagu Rangan. 2024. Tnt-llm: Text mining at scale with large language models. Preprint, arXiv:2403.12173. — [arXiv:2403.12173](https://arxiv.org/abs/2403.12173) find
  19. Qingkai Zeng, Yuyang Bai, Zhaoxuan Tan, Shangbin Feng, Zhenwen Liang, Zhihan Zhang, and Meng Jiang. 2024. Chain-of-layer: Iteratively prompting large language models for taxonomy induction from limited examples. Preprint, arXiv:2402.07386. — [arXiv:2402.07386](https://arxiv.org/abs/2402.07386) find
  20. Hanlei Zhang, Hua Xu, Xin Wang, Fei Long, and Kai Gao. 2024. A clustering framework for unsupervised and semi-supervised new intent discovery. IEEE Transactions on Knowledge and Data Engineering, 36(11):5468–5481. find
  21. Kun Zhu, Lizi Liao, Yuxuan Gu, Lei Huang, Xiaocheng Feng, and Bing Qin. 2025. Context-aware hierarchical taxonomy generation for scientific papers via LLM-guided multi-aspect clustering. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 15616–15634, Suz… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mantri-2026-hierarchical-reason-contact-detection-retail/